xref: /linux/arch/powerpc/kernel/setup_64.c (revision d195c39052d1da278a00a6744ce59c383b67b191)
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 __nostackprotector 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 	/*
289 	 * Assume we're on cpu 0 for now.
290 	 *
291 	 * We need to load a PACA very early for a few reasons.
292 	 *
293 	 * The stack protector canary is stored in the paca, so as soon as we
294 	 * call any stack protected code we need r13 pointing somewhere valid.
295 	 *
296 	 * If we are using kcov it will call in_task() in its instrumentation,
297 	 * which relies on the current task from the PACA.
298 	 *
299 	 * dt_cpu_ftrs_init() calls into generic OF/fdt code, as well as
300 	 * printk(), which can trigger both stack protector and kcov.
301 	 *
302 	 * percpu variables and spin locks also use the paca.
303 	 *
304 	 * So set up a temporary paca. It will be replaced below once we know
305 	 * what CPU we are on.
306 	 */
307 	initialise_paca(&boot_paca, 0);
308 	setup_paca(&boot_paca);
309 	fixup_boot_paca();
310 
311 	/* -------- printk is now safe to use ------- */
312 
313 	/* Try new device tree based feature discovery ... */
314 	if (!dt_cpu_ftrs_init(__va(dt_ptr)))
315 		/* Otherwise use the old style CPU table */
316 		identify_cpu(0, mfspr(SPRN_PVR));
317 
318 	/* Enable early debugging if any specified (see udbg.h) */
319 	udbg_early_init();
320 
321 	udbg_printf(" -> %s(), dt_ptr: 0x%lx\n", __func__, dt_ptr);
322 
323 	/*
324 	 * Do early initialization using the flattened device
325 	 * tree, such as retrieving the physical memory map or
326 	 * calculating/retrieving the hash table size.
327 	 */
328 	early_init_devtree(__va(dt_ptr));
329 
330 	/* Now we know the logical id of our boot cpu, setup the paca. */
331 	if (boot_cpuid != 0) {
332 		/* Poison paca_ptrs[0] again if it's not the boot cpu */
333 		memset(&paca_ptrs[0], 0x88, sizeof(paca_ptrs[0]));
334 	}
335 	setup_paca(paca_ptrs[boot_cpuid]);
336 	fixup_boot_paca();
337 
338 	/*
339 	 * Configure exception handlers. This include setting up trampolines
340 	 * if needed, setting exception endian mode, etc...
341 	 */
342 	configure_exceptions();
343 
344 	/*
345 	 * Configure Kernel Userspace Protection. This needs to happen before
346 	 * feature fixups for platforms that implement this using features.
347 	 */
348 	setup_kup();
349 
350 	/* Apply all the dynamic patching */
351 	apply_feature_fixups();
352 	setup_feature_keys();
353 
354 	early_ioremap_setup();
355 
356 	/* Initialize the hash table or TLB handling */
357 	early_init_mmu();
358 
359 	/*
360 	 * After firmware and early platform setup code has set things up,
361 	 * we note the SPR values for configurable control/performance
362 	 * registers, and use those as initial defaults.
363 	 */
364 	record_spr_defaults();
365 
366 	/*
367 	 * At this point, we can let interrupts switch to virtual mode
368 	 * (the MMU has been setup), so adjust the MSR in the PACA to
369 	 * have IR and DR set and enable AIL if it exists
370 	 */
371 	cpu_ready_for_interrupts();
372 
373 	/*
374 	 * We enable ftrace here, but since we only support DYNAMIC_FTRACE, it
375 	 * will only actually get enabled on the boot cpu much later once
376 	 * ftrace itself has been initialized.
377 	 */
378 	this_cpu_enable_ftrace();
379 
380 	udbg_printf(" <- %s()\n", __func__);
381 
382 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
383 	/*
384 	 * This needs to be done *last* (after the above udbg_printf() even)
385 	 *
386 	 * Right after we return from this function, we turn on the MMU
387 	 * which means the real-mode access trick that btext does will
388 	 * no longer work, it needs to switch to using a real MMU
389 	 * mapping. This call will ensure that it does
390 	 */
391 	btext_map();
392 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
393 }
394 
395 #ifdef CONFIG_SMP
396 void early_setup_secondary(void)
397 {
398 	/* Mark interrupts disabled in PACA */
399 	irq_soft_mask_set(IRQS_DISABLED);
400 
401 	/* Initialize the hash table or TLB handling */
402 	early_init_mmu_secondary();
403 
404 	/* Perform any KUP setup that is per-cpu */
405 	setup_kup();
406 
407 	/*
408 	 * At this point, we can let interrupts switch to virtual mode
409 	 * (the MMU has been setup), so adjust the MSR in the PACA to
410 	 * have IR and DR set.
411 	 */
412 	cpu_ready_for_interrupts();
413 }
414 
415 #endif /* CONFIG_SMP */
416 
417 void panic_smp_self_stop(void)
418 {
419 	hard_irq_disable();
420 	spin_begin();
421 	while (1)
422 		spin_cpu_relax();
423 }
424 
425 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC_CORE)
426 static bool use_spinloop(void)
427 {
428 	if (IS_ENABLED(CONFIG_PPC_BOOK3S)) {
429 		/*
430 		 * See comments in head_64.S -- not all platforms insert
431 		 * secondaries at __secondary_hold and wait at the spin
432 		 * loop.
433 		 */
434 		if (firmware_has_feature(FW_FEATURE_OPAL))
435 			return false;
436 		return true;
437 	}
438 
439 	/*
440 	 * When book3e boots from kexec, the ePAPR spin table does
441 	 * not get used.
442 	 */
443 	return of_property_read_bool(of_chosen, "linux,booted-from-kexec");
444 }
445 
446 void smp_release_cpus(void)
447 {
448 	unsigned long *ptr;
449 	int i;
450 
451 	if (!use_spinloop())
452 		return;
453 
454 	/* All secondary cpus are spinning on a common spinloop, release them
455 	 * all now so they can start to spin on their individual paca
456 	 * spinloops. For non SMP kernels, the secondary cpus never get out
457 	 * of the common spinloop.
458 	 */
459 
460 	ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
461 			- PHYSICAL_START);
462 	*ptr = ppc_function_entry(generic_secondary_smp_init);
463 
464 	/* And wait a bit for them to catch up */
465 	for (i = 0; i < 100000; i++) {
466 		mb();
467 		HMT_low();
468 		if (spinning_secondaries == 0)
469 			break;
470 		udelay(1);
471 	}
472 	pr_debug("spinning_secondaries = %d\n", spinning_secondaries);
473 }
474 #endif /* CONFIG_SMP || CONFIG_KEXEC_CORE */
475 
476 /*
477  * Initialize some remaining members of the ppc64_caches and systemcfg
478  * structures
479  * (at least until we get rid of them completely). This is mostly some
480  * cache informations about the CPU that will be used by cache flush
481  * routines and/or provided to userland
482  */
483 
484 static void init_cache_info(struct ppc_cache_info *info, u32 size, u32 lsize,
485 			    u32 bsize, u32 sets)
486 {
487 	info->size = size;
488 	info->sets = sets;
489 	info->line_size = lsize;
490 	info->block_size = bsize;
491 	info->log_block_size = __ilog2(bsize);
492 	if (bsize)
493 		info->blocks_per_page = PAGE_SIZE / bsize;
494 	else
495 		info->blocks_per_page = 0;
496 
497 	if (sets == 0)
498 		info->assoc = 0xffff;
499 	else
500 		info->assoc = size / (sets * lsize);
501 }
502 
503 static bool __init parse_cache_info(struct device_node *np,
504 				    bool icache,
505 				    struct ppc_cache_info *info)
506 {
507 	static const char *ipropnames[] __initdata = {
508 		"i-cache-size",
509 		"i-cache-sets",
510 		"i-cache-block-size",
511 		"i-cache-line-size",
512 	};
513 	static const char *dpropnames[] __initdata = {
514 		"d-cache-size",
515 		"d-cache-sets",
516 		"d-cache-block-size",
517 		"d-cache-line-size",
518 	};
519 	const char **propnames = icache ? ipropnames : dpropnames;
520 	const __be32 *sizep, *lsizep, *bsizep, *setsp;
521 	u32 size, lsize, bsize, sets;
522 	bool success = true;
523 
524 	size = 0;
525 	sets = -1u;
526 	lsize = bsize = cur_cpu_spec->dcache_bsize;
527 	sizep = of_get_property(np, propnames[0], NULL);
528 	if (sizep != NULL)
529 		size = be32_to_cpu(*sizep);
530 	setsp = of_get_property(np, propnames[1], NULL);
531 	if (setsp != NULL)
532 		sets = be32_to_cpu(*setsp);
533 	bsizep = of_get_property(np, propnames[2], NULL);
534 	lsizep = of_get_property(np, propnames[3], NULL);
535 	if (bsizep == NULL)
536 		bsizep = lsizep;
537 	if (lsizep != NULL)
538 		lsize = be32_to_cpu(*lsizep);
539 	if (bsizep != NULL)
540 		bsize = be32_to_cpu(*bsizep);
541 	if (sizep == NULL || bsizep == NULL || lsizep == NULL)
542 		success = false;
543 
544 	/*
545 	 * OF is weird .. it represents fully associative caches
546 	 * as "1 way" which doesn't make much sense and doesn't
547 	 * leave room for direct mapped. We'll assume that 0
548 	 * in OF means direct mapped for that reason.
549 	 */
550 	if (sets == 1)
551 		sets = 0;
552 	else if (sets == 0)
553 		sets = 1;
554 
555 	init_cache_info(info, size, lsize, bsize, sets);
556 
557 	return success;
558 }
559 
560 void __init initialize_cache_info(void)
561 {
562 	struct device_node *cpu = NULL, *l2, *l3 = NULL;
563 	u32 pvr;
564 
565 	/*
566 	 * All shipping POWER8 machines have a firmware bug that
567 	 * puts incorrect information in the device-tree. This will
568 	 * be (hopefully) fixed for future chips but for now hard
569 	 * code the values if we are running on one of these
570 	 */
571 	pvr = PVR_VER(mfspr(SPRN_PVR));
572 	if (pvr == PVR_POWER8 || pvr == PVR_POWER8E ||
573 	    pvr == PVR_POWER8NVL) {
574 						/* size    lsize   blk  sets */
575 		init_cache_info(&ppc64_caches.l1i, 0x8000,   128,  128, 32);
576 		init_cache_info(&ppc64_caches.l1d, 0x10000,  128,  128, 64);
577 		init_cache_info(&ppc64_caches.l2,  0x80000,  128,  0,   512);
578 		init_cache_info(&ppc64_caches.l3,  0x800000, 128,  0,   8192);
579 	} else
580 		cpu = of_find_node_by_type(NULL, "cpu");
581 
582 	/*
583 	 * We're assuming *all* of the CPUs have the same
584 	 * d-cache and i-cache sizes... -Peter
585 	 */
586 	if (cpu) {
587 		if (!parse_cache_info(cpu, false, &ppc64_caches.l1d))
588 			pr_warn("Argh, can't find dcache properties !\n");
589 
590 		if (!parse_cache_info(cpu, true, &ppc64_caches.l1i))
591 			pr_warn("Argh, can't find icache properties !\n");
592 
593 		/*
594 		 * Try to find the L2 and L3 if any. Assume they are
595 		 * unified and use the D-side properties.
596 		 */
597 		l2 = of_find_next_cache_node(cpu);
598 		of_node_put(cpu);
599 		if (l2) {
600 			parse_cache_info(l2, false, &ppc64_caches.l2);
601 			l3 = of_find_next_cache_node(l2);
602 			of_node_put(l2);
603 		}
604 		if (l3) {
605 			parse_cache_info(l3, false, &ppc64_caches.l3);
606 			of_node_put(l3);
607 		}
608 	}
609 
610 	/* For use by binfmt_elf */
611 	dcache_bsize = ppc64_caches.l1d.block_size;
612 	icache_bsize = ppc64_caches.l1i.block_size;
613 
614 	cur_cpu_spec->dcache_bsize = dcache_bsize;
615 	cur_cpu_spec->icache_bsize = icache_bsize;
616 }
617 
618 /*
619  * This returns the limit below which memory accesses to the linear
620  * mapping are guarnateed not to cause an architectural exception (e.g.,
621  * TLB or SLB miss fault).
622  *
623  * This is used to allocate PACAs and various interrupt stacks that
624  * that are accessed early in interrupt handlers that must not cause
625  * re-entrant interrupts.
626  */
627 __init u64 ppc64_bolted_size(void)
628 {
629 #ifdef CONFIG_PPC_BOOK3E
630 	/* Freescale BookE bolts the entire linear mapping */
631 	/* XXX: BookE ppc64_rma_limit setup seems to disagree? */
632 	if (early_mmu_has_feature(MMU_FTR_TYPE_FSL_E))
633 		return linear_map_top;
634 	/* Other BookE, we assume the first GB is bolted */
635 	return 1ul << 30;
636 #else
637 	/* BookS radix, does not take faults on linear mapping */
638 	if (early_radix_enabled())
639 		return ULONG_MAX;
640 
641 	/* BookS hash, the first segment is bolted */
642 	if (early_mmu_has_feature(MMU_FTR_1T_SEGMENT))
643 		return 1UL << SID_SHIFT_1T;
644 	return 1UL << SID_SHIFT;
645 #endif
646 }
647 
648 static void *__init alloc_stack(unsigned long limit, int cpu)
649 {
650 	void *ptr;
651 
652 	BUILD_BUG_ON(STACK_INT_FRAME_SIZE % 16);
653 
654 	ptr = memblock_alloc_try_nid(THREAD_SIZE, THREAD_ALIGN,
655 				     MEMBLOCK_LOW_LIMIT, limit,
656 				     early_cpu_to_node(cpu));
657 	if (!ptr)
658 		panic("cannot allocate stacks");
659 
660 	return ptr;
661 }
662 
663 void __init irqstack_early_init(void)
664 {
665 	u64 limit = ppc64_bolted_size();
666 	unsigned int i;
667 
668 	/*
669 	 * Interrupt stacks must be in the first segment since we
670 	 * cannot afford to take SLB misses on them. They are not
671 	 * accessed in realmode.
672 	 */
673 	for_each_possible_cpu(i) {
674 		softirq_ctx[i] = alloc_stack(limit, i);
675 		hardirq_ctx[i] = alloc_stack(limit, i);
676 	}
677 }
678 
679 #ifdef CONFIG_PPC_BOOK3E
680 void __init exc_lvl_early_init(void)
681 {
682 	unsigned int i;
683 
684 	for_each_possible_cpu(i) {
685 		void *sp;
686 
687 		sp = alloc_stack(ULONG_MAX, i);
688 		critirq_ctx[i] = sp;
689 		paca_ptrs[i]->crit_kstack = sp + THREAD_SIZE;
690 
691 		sp = alloc_stack(ULONG_MAX, i);
692 		dbgirq_ctx[i] = sp;
693 		paca_ptrs[i]->dbg_kstack = sp + THREAD_SIZE;
694 
695 		sp = alloc_stack(ULONG_MAX, i);
696 		mcheckirq_ctx[i] = sp;
697 		paca_ptrs[i]->mc_kstack = sp + THREAD_SIZE;
698 	}
699 
700 	if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC))
701 		patch_exception(0x040, exc_debug_debug_book3e);
702 }
703 #endif
704 
705 /*
706  * Stack space used when we detect a bad kernel stack pointer, and
707  * early in SMP boots before relocation is enabled. Exclusive emergency
708  * stack for machine checks.
709  */
710 void __init emergency_stack_init(void)
711 {
712 	u64 limit;
713 	unsigned int i;
714 
715 	/*
716 	 * Emergency stacks must be under 256MB, we cannot afford to take
717 	 * SLB misses on them. The ABI also requires them to be 128-byte
718 	 * aligned.
719 	 *
720 	 * Since we use these as temporary stacks during secondary CPU
721 	 * bringup, machine check, system reset, and HMI, we need to get
722 	 * at them in real mode. This means they must also be within the RMO
723 	 * region.
724 	 *
725 	 * The IRQ stacks allocated elsewhere in this file are zeroed and
726 	 * initialized in kernel/irq.c. These are initialized here in order
727 	 * to have emergency stacks available as early as possible.
728 	 */
729 	limit = min(ppc64_bolted_size(), ppc64_rma_size);
730 
731 	for_each_possible_cpu(i) {
732 		paca_ptrs[i]->emergency_sp = alloc_stack(limit, i) + THREAD_SIZE;
733 
734 #ifdef CONFIG_PPC_BOOK3S_64
735 		/* emergency stack for NMI exception handling. */
736 		paca_ptrs[i]->nmi_emergency_sp = alloc_stack(limit, i) + THREAD_SIZE;
737 
738 		/* emergency stack for machine check exception handling. */
739 		paca_ptrs[i]->mc_emergency_sp = alloc_stack(limit, i) + THREAD_SIZE;
740 #endif
741 	}
742 }
743 
744 #ifdef CONFIG_SMP
745 #define PCPU_DYN_SIZE		()
746 
747 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
748 {
749 	return memblock_alloc_try_nid(size, align, __pa(MAX_DMA_ADDRESS),
750 				      MEMBLOCK_ALLOC_ACCESSIBLE,
751 				      early_cpu_to_node(cpu));
752 
753 }
754 
755 static void __init pcpu_fc_free(void *ptr, size_t size)
756 {
757 	memblock_free(__pa(ptr), size);
758 }
759 
760 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
761 {
762 	if (early_cpu_to_node(from) == early_cpu_to_node(to))
763 		return LOCAL_DISTANCE;
764 	else
765 		return REMOTE_DISTANCE;
766 }
767 
768 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
769 EXPORT_SYMBOL(__per_cpu_offset);
770 
771 void __init setup_per_cpu_areas(void)
772 {
773 	const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
774 	size_t atom_size;
775 	unsigned long delta;
776 	unsigned int cpu;
777 	int rc;
778 
779 	/*
780 	 * Linear mapping is one of 4K, 1M and 16M.  For 4K, no need
781 	 * to group units.  For larger mappings, use 1M atom which
782 	 * should be large enough to contain a number of units.
783 	 */
784 	if (mmu_linear_psize == MMU_PAGE_4K)
785 		atom_size = PAGE_SIZE;
786 	else
787 		atom_size = 1 << 20;
788 
789 	rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
790 				    pcpu_fc_alloc, pcpu_fc_free);
791 	if (rc < 0)
792 		panic("cannot initialize percpu area (err=%d)", rc);
793 
794 	delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
795 	for_each_possible_cpu(cpu) {
796                 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
797 		paca_ptrs[cpu]->data_offset = __per_cpu_offset[cpu];
798 	}
799 }
800 #endif
801 
802 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
803 unsigned long memory_block_size_bytes(void)
804 {
805 	if (ppc_md.memory_block_size)
806 		return ppc_md.memory_block_size();
807 
808 	return MIN_MEMORY_BLOCK_SIZE;
809 }
810 #endif
811 
812 #if defined(CONFIG_PPC_INDIRECT_PIO) || defined(CONFIG_PPC_INDIRECT_MMIO)
813 struct ppc_pci_io ppc_pci_io;
814 EXPORT_SYMBOL(ppc_pci_io);
815 #endif
816 
817 #ifdef CONFIG_HARDLOCKUP_DETECTOR_PERF
818 u64 hw_nmi_get_sample_period(int watchdog_thresh)
819 {
820 	return ppc_proc_freq * watchdog_thresh;
821 }
822 #endif
823 
824 /*
825  * The perf based hardlockup detector breaks PMU event based branches, so
826  * disable it by default. Book3S has a soft-nmi hardlockup detector based
827  * on the decrementer interrupt, so it does not suffer from this problem.
828  *
829  * It is likely to get false positives in VM guests, so disable it there
830  * by default too.
831  */
832 static int __init disable_hardlockup_detector(void)
833 {
834 #ifdef CONFIG_HARDLOCKUP_DETECTOR_PERF
835 	hardlockup_detector_disable();
836 #else
837 	if (firmware_has_feature(FW_FEATURE_LPAR))
838 		hardlockup_detector_disable();
839 #endif
840 
841 	return 0;
842 }
843 early_initcall(disable_hardlockup_detector);
844 
845 #ifdef CONFIG_PPC_BOOK3S_64
846 static enum l1d_flush_type enabled_flush_types;
847 static void *l1d_flush_fallback_area;
848 static bool no_rfi_flush;
849 bool rfi_flush;
850 
851 static int __init handle_no_rfi_flush(char *p)
852 {
853 	pr_info("rfi-flush: disabled on command line.");
854 	no_rfi_flush = true;
855 	return 0;
856 }
857 early_param("no_rfi_flush", handle_no_rfi_flush);
858 
859 /*
860  * The RFI flush is not KPTI, but because users will see doco that says to use
861  * nopti we hijack that option here to also disable the RFI flush.
862  */
863 static int __init handle_no_pti(char *p)
864 {
865 	pr_info("rfi-flush: disabling due to 'nopti' on command line.\n");
866 	handle_no_rfi_flush(NULL);
867 	return 0;
868 }
869 early_param("nopti", handle_no_pti);
870 
871 static void do_nothing(void *unused)
872 {
873 	/*
874 	 * We don't need to do the flush explicitly, just enter+exit kernel is
875 	 * sufficient, the RFI exit handlers will do the right thing.
876 	 */
877 }
878 
879 void rfi_flush_enable(bool enable)
880 {
881 	if (enable) {
882 		do_rfi_flush_fixups(enabled_flush_types);
883 		on_each_cpu(do_nothing, NULL, 1);
884 	} else
885 		do_rfi_flush_fixups(L1D_FLUSH_NONE);
886 
887 	rfi_flush = enable;
888 }
889 
890 static void __ref init_fallback_flush(void)
891 {
892 	u64 l1d_size, limit;
893 	int cpu;
894 
895 	/* Only allocate the fallback flush area once (at boot time). */
896 	if (l1d_flush_fallback_area)
897 		return;
898 
899 	l1d_size = ppc64_caches.l1d.size;
900 
901 	/*
902 	 * If there is no d-cache-size property in the device tree, l1d_size
903 	 * could be zero. That leads to the loop in the asm wrapping around to
904 	 * 2^64-1, and then walking off the end of the fallback area and
905 	 * eventually causing a page fault which is fatal. Just default to
906 	 * something vaguely sane.
907 	 */
908 	if (!l1d_size)
909 		l1d_size = (64 * 1024);
910 
911 	limit = min(ppc64_bolted_size(), ppc64_rma_size);
912 
913 	/*
914 	 * Align to L1d size, and size it at 2x L1d size, to catch possible
915 	 * hardware prefetch runoff. We don't have a recipe for load patterns to
916 	 * reliably avoid the prefetcher.
917 	 */
918 	l1d_flush_fallback_area = memblock_alloc_try_nid(l1d_size * 2,
919 						l1d_size, MEMBLOCK_LOW_LIMIT,
920 						limit, NUMA_NO_NODE);
921 	if (!l1d_flush_fallback_area)
922 		panic("%s: Failed to allocate %llu bytes align=0x%llx max_addr=%pa\n",
923 		      __func__, l1d_size * 2, l1d_size, &limit);
924 
925 
926 	for_each_possible_cpu(cpu) {
927 		struct paca_struct *paca = paca_ptrs[cpu];
928 		paca->rfi_flush_fallback_area = l1d_flush_fallback_area;
929 		paca->l1d_flush_size = l1d_size;
930 	}
931 }
932 
933 void setup_rfi_flush(enum l1d_flush_type types, bool enable)
934 {
935 	if (types & L1D_FLUSH_FALLBACK) {
936 		pr_info("rfi-flush: fallback displacement flush available\n");
937 		init_fallback_flush();
938 	}
939 
940 	if (types & L1D_FLUSH_ORI)
941 		pr_info("rfi-flush: ori type flush available\n");
942 
943 	if (types & L1D_FLUSH_MTTRIG)
944 		pr_info("rfi-flush: mttrig type flush available\n");
945 
946 	enabled_flush_types = types;
947 
948 	if (!no_rfi_flush && !cpu_mitigations_off())
949 		rfi_flush_enable(enable);
950 }
951 
952 #ifdef CONFIG_DEBUG_FS
953 static int rfi_flush_set(void *data, u64 val)
954 {
955 	bool enable;
956 
957 	if (val == 1)
958 		enable = true;
959 	else if (val == 0)
960 		enable = false;
961 	else
962 		return -EINVAL;
963 
964 	/* Only do anything if we're changing state */
965 	if (enable != rfi_flush)
966 		rfi_flush_enable(enable);
967 
968 	return 0;
969 }
970 
971 static int rfi_flush_get(void *data, u64 *val)
972 {
973 	*val = rfi_flush ? 1 : 0;
974 	return 0;
975 }
976 
977 DEFINE_SIMPLE_ATTRIBUTE(fops_rfi_flush, rfi_flush_get, rfi_flush_set, "%llu\n");
978 
979 static __init int rfi_flush_debugfs_init(void)
980 {
981 	debugfs_create_file("rfi_flush", 0600, powerpc_debugfs_root, NULL, &fops_rfi_flush);
982 	return 0;
983 }
984 device_initcall(rfi_flush_debugfs_init);
985 #endif
986 #endif /* CONFIG_PPC_BOOK3S_64 */
987