xref: /linux/arch/powerpc/kernel/setup_64.c (revision 140eb5227767c6754742020a16d2691222b9c19b)
1 /*
2  *
3  * Common boot and setup code.
4  *
5  * Copyright (C) 2001 PPC64 Team, IBM Corp
6  *
7  *      This program is free software; you can redistribute it and/or
8  *      modify it under the terms of the GNU General Public License
9  *      as published by the Free Software Foundation; either version
10  *      2 of the License, or (at your option) any later version.
11  */
12 
13 #define DEBUG
14 
15 #include <linux/export.h>
16 #include <linux/string.h>
17 #include <linux/sched.h>
18 #include <linux/init.h>
19 #include <linux/kernel.h>
20 #include <linux/reboot.h>
21 #include <linux/delay.h>
22 #include <linux/initrd.h>
23 #include <linux/seq_file.h>
24 #include <linux/ioport.h>
25 #include <linux/console.h>
26 #include <linux/utsname.h>
27 #include <linux/tty.h>
28 #include <linux/root_dev.h>
29 #include <linux/notifier.h>
30 #include <linux/cpu.h>
31 #include <linux/unistd.h>
32 #include <linux/serial.h>
33 #include <linux/serial_8250.h>
34 #include <linux/bootmem.h>
35 #include <linux/pci.h>
36 #include <linux/lockdep.h>
37 #include <linux/memblock.h>
38 #include <linux/memory.h>
39 #include <linux/nmi.h>
40 
41 #include <asm/debugfs.h>
42 #include <asm/io.h>
43 #include <asm/kdump.h>
44 #include <asm/prom.h>
45 #include <asm/processor.h>
46 #include <asm/pgtable.h>
47 #include <asm/smp.h>
48 #include <asm/elf.h>
49 #include <asm/machdep.h>
50 #include <asm/paca.h>
51 #include <asm/time.h>
52 #include <asm/cputable.h>
53 #include <asm/dt_cpu_ftrs.h>
54 #include <asm/sections.h>
55 #include <asm/btext.h>
56 #include <asm/nvram.h>
57 #include <asm/setup.h>
58 #include <asm/rtas.h>
59 #include <asm/iommu.h>
60 #include <asm/serial.h>
61 #include <asm/cache.h>
62 #include <asm/page.h>
63 #include <asm/mmu.h>
64 #include <asm/firmware.h>
65 #include <asm/xmon.h>
66 #include <asm/udbg.h>
67 #include <asm/kexec.h>
68 #include <asm/code-patching.h>
69 #include <asm/livepatch.h>
70 #include <asm/opal.h>
71 #include <asm/cputhreads.h>
72 
73 #include "setup.h"
74 
75 #ifdef DEBUG
76 #define DBG(fmt...) udbg_printf(fmt)
77 #else
78 #define DBG(fmt...)
79 #endif
80 
81 int spinning_secondaries;
82 u64 ppc64_pft_size;
83 
84 struct ppc64_caches ppc64_caches = {
85 	.l1d = {
86 		.block_size = 0x40,
87 		.log_block_size = 6,
88 	},
89 	.l1i = {
90 		.block_size = 0x40,
91 		.log_block_size = 6
92 	},
93 };
94 EXPORT_SYMBOL_GPL(ppc64_caches);
95 
96 #if defined(CONFIG_PPC_BOOK3E) && defined(CONFIG_SMP)
97 void __init setup_tlb_core_data(void)
98 {
99 	int cpu;
100 
101 	BUILD_BUG_ON(offsetof(struct tlb_core_data, lock) != 0);
102 
103 	for_each_possible_cpu(cpu) {
104 		int first = cpu_first_thread_sibling(cpu);
105 
106 		/*
107 		 * If we boot via kdump on a non-primary thread,
108 		 * make sure we point at the thread that actually
109 		 * set up this TLB.
110 		 */
111 		if (cpu_first_thread_sibling(boot_cpuid) == first)
112 			first = boot_cpuid;
113 
114 		paca[cpu].tcd_ptr = &paca[first].tcd;
115 
116 		/*
117 		 * If we have threads, we need either tlbsrx.
118 		 * or e6500 tablewalk mode, or else TLB handlers
119 		 * will be racy and could produce duplicate entries.
120 		 * Should we panic instead?
121 		 */
122 		WARN_ONCE(smt_enabled_at_boot >= 2 &&
123 			  !mmu_has_feature(MMU_FTR_USE_TLBRSRV) &&
124 			  book3e_htw_mode != PPC_HTW_E6500,
125 			  "%s: unsupported MMU configuration\n", __func__);
126 	}
127 }
128 #endif
129 
130 #ifdef CONFIG_SMP
131 
132 static char *smt_enabled_cmdline;
133 
134 /* Look for ibm,smt-enabled OF option */
135 void __init check_smt_enabled(void)
136 {
137 	struct device_node *dn;
138 	const char *smt_option;
139 
140 	/* Default to enabling all threads */
141 	smt_enabled_at_boot = threads_per_core;
142 
143 	/* Allow the command line to overrule the OF option */
144 	if (smt_enabled_cmdline) {
145 		if (!strcmp(smt_enabled_cmdline, "on"))
146 			smt_enabled_at_boot = threads_per_core;
147 		else if (!strcmp(smt_enabled_cmdline, "off"))
148 			smt_enabled_at_boot = 0;
149 		else {
150 			int smt;
151 			int rc;
152 
153 			rc = kstrtoint(smt_enabled_cmdline, 10, &smt);
154 			if (!rc)
155 				smt_enabled_at_boot =
156 					min(threads_per_core, smt);
157 		}
158 	} else {
159 		dn = of_find_node_by_path("/options");
160 		if (dn) {
161 			smt_option = of_get_property(dn, "ibm,smt-enabled",
162 						     NULL);
163 
164 			if (smt_option) {
165 				if (!strcmp(smt_option, "on"))
166 					smt_enabled_at_boot = threads_per_core;
167 				else if (!strcmp(smt_option, "off"))
168 					smt_enabled_at_boot = 0;
169 			}
170 
171 			of_node_put(dn);
172 		}
173 	}
174 }
175 
176 /* Look for smt-enabled= cmdline option */
177 static int __init early_smt_enabled(char *p)
178 {
179 	smt_enabled_cmdline = p;
180 	return 0;
181 }
182 early_param("smt-enabled", early_smt_enabled);
183 
184 #endif /* CONFIG_SMP */
185 
186 /** Fix up paca fields required for the boot cpu */
187 static void __init fixup_boot_paca(void)
188 {
189 	/* The boot cpu is started */
190 	get_paca()->cpu_start = 1;
191 	/* Allow percpu accesses to work until we setup percpu data */
192 	get_paca()->data_offset = 0;
193 }
194 
195 static void __init configure_exceptions(void)
196 {
197 	/*
198 	 * Setup the trampolines from the lowmem exception vectors
199 	 * to the kdump kernel when not using a relocatable kernel.
200 	 */
201 	setup_kdump_trampoline();
202 
203 	/* Under a PAPR hypervisor, we need hypercalls */
204 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
205 		/* Enable AIL if possible */
206 		pseries_enable_reloc_on_exc();
207 
208 		/*
209 		 * Tell the hypervisor that we want our exceptions to
210 		 * be taken in little endian mode.
211 		 *
212 		 * We don't call this for big endian as our calling convention
213 		 * makes us always enter in BE, and the call may fail under
214 		 * some circumstances with kdump.
215 		 */
216 #ifdef __LITTLE_ENDIAN__
217 		pseries_little_endian_exceptions();
218 #endif
219 	} else {
220 		/* Set endian mode using OPAL */
221 		if (firmware_has_feature(FW_FEATURE_OPAL))
222 			opal_configure_cores();
223 
224 		/* AIL on native is done in cpu_ready_for_interrupts() */
225 	}
226 }
227 
228 static void cpu_ready_for_interrupts(void)
229 {
230 	/*
231 	 * Enable AIL if supported, and we are in hypervisor mode. This
232 	 * is called once for every processor.
233 	 *
234 	 * If we are not in hypervisor mode the job is done once for
235 	 * the whole partition in configure_exceptions().
236 	 */
237 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
238 	    cpu_has_feature(CPU_FTR_ARCH_207S)) {
239 		unsigned long lpcr = mfspr(SPRN_LPCR);
240 		mtspr(SPRN_LPCR, lpcr | LPCR_AIL_3);
241 	}
242 
243 	/*
244 	 * Fixup HFSCR:TM based on CPU features. The bit is set by our
245 	 * early asm init because at that point we haven't updated our
246 	 * CPU features from firmware and device-tree. Here we have,
247 	 * so let's do it.
248 	 */
249 	if (cpu_has_feature(CPU_FTR_HVMODE) && !cpu_has_feature(CPU_FTR_TM_COMP))
250 		mtspr(SPRN_HFSCR, mfspr(SPRN_HFSCR) & ~HFSCR_TM);
251 
252 	/* Set IR and DR in PACA MSR */
253 	get_paca()->kernel_msr = MSR_KERNEL;
254 }
255 
256 /*
257  * Early initialization entry point. This is called by head.S
258  * with MMU translation disabled. We rely on the "feature" of
259  * the CPU that ignores the top 2 bits of the address in real
260  * mode so we can access kernel globals normally provided we
261  * only toy with things in the RMO region. From here, we do
262  * some early parsing of the device-tree to setup out MEMBLOCK
263  * data structures, and allocate & initialize the hash table
264  * and segment tables so we can start running with translation
265  * enabled.
266  *
267  * It is this function which will call the probe() callback of
268  * the various platform types and copy the matching one to the
269  * global ppc_md structure. Your platform can eventually do
270  * some very early initializations from the probe() routine, but
271  * this is not recommended, be very careful as, for example, the
272  * device-tree is not accessible via normal means at this point.
273  */
274 
275 void __init early_setup(unsigned long dt_ptr)
276 {
277 	static __initdata struct paca_struct boot_paca;
278 
279 	/* -------- printk is _NOT_ safe to use here ! ------- */
280 
281 	/* Try new device tree based feature discovery ... */
282 	if (!dt_cpu_ftrs_init(__va(dt_ptr)))
283 		/* Otherwise use the old style CPU table */
284 		identify_cpu(0, mfspr(SPRN_PVR));
285 
286 	/* Assume we're on cpu 0 for now. Don't write to the paca yet! */
287 	initialise_paca(&boot_paca, 0);
288 	setup_paca(&boot_paca);
289 	fixup_boot_paca();
290 
291 	/* -------- printk is now safe to use ------- */
292 
293 	/* Enable early debugging if any specified (see udbg.h) */
294 	udbg_early_init();
295 
296  	DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
297 
298 	/*
299 	 * Do early initialization using the flattened device
300 	 * tree, such as retrieving the physical memory map or
301 	 * calculating/retrieving the hash table size.
302 	 */
303 	early_init_devtree(__va(dt_ptr));
304 
305 	/* Now we know the logical id of our boot cpu, setup the paca. */
306 	setup_paca(&paca[boot_cpuid]);
307 	fixup_boot_paca();
308 
309 	/*
310 	 * Configure exception handlers. This include setting up trampolines
311 	 * if needed, setting exception endian mode, etc...
312 	 */
313 	configure_exceptions();
314 
315 	/* Apply all the dynamic patching */
316 	apply_feature_fixups();
317 	setup_feature_keys();
318 
319 	/* Initialize the hash table or TLB handling */
320 	early_init_mmu();
321 
322 	/*
323 	 * After firmware and early platform setup code has set things up,
324 	 * we note the SPR values for configurable control/performance
325 	 * registers, and use those as initial defaults.
326 	 */
327 	record_spr_defaults();
328 
329 	/*
330 	 * At this point, we can let interrupts switch to virtual mode
331 	 * (the MMU has been setup), so adjust the MSR in the PACA to
332 	 * have IR and DR set and enable AIL if it exists
333 	 */
334 	cpu_ready_for_interrupts();
335 
336 	DBG(" <- early_setup()\n");
337 
338 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
339 	/*
340 	 * This needs to be done *last* (after the above DBG() even)
341 	 *
342 	 * Right after we return from this function, we turn on the MMU
343 	 * which means the real-mode access trick that btext does will
344 	 * no longer work, it needs to switch to using a real MMU
345 	 * mapping. This call will ensure that it does
346 	 */
347 	btext_map();
348 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
349 }
350 
351 #ifdef CONFIG_SMP
352 void early_setup_secondary(void)
353 {
354 	/* Mark interrupts disabled in PACA */
355 	get_paca()->soft_enabled = 0;
356 
357 	/* Initialize the hash table or TLB handling */
358 	early_init_mmu_secondary();
359 
360 	/*
361 	 * At this point, we can let interrupts switch to virtual mode
362 	 * (the MMU has been setup), so adjust the MSR in the PACA to
363 	 * have IR and DR set.
364 	 */
365 	cpu_ready_for_interrupts();
366 }
367 
368 #endif /* CONFIG_SMP */
369 
370 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC_CORE)
371 static bool use_spinloop(void)
372 {
373 	if (IS_ENABLED(CONFIG_PPC_BOOK3S)) {
374 		/*
375 		 * See comments in head_64.S -- not all platforms insert
376 		 * secondaries at __secondary_hold and wait at the spin
377 		 * loop.
378 		 */
379 		if (firmware_has_feature(FW_FEATURE_OPAL))
380 			return false;
381 		return true;
382 	}
383 
384 	/*
385 	 * When book3e boots from kexec, the ePAPR spin table does
386 	 * not get used.
387 	 */
388 	return of_property_read_bool(of_chosen, "linux,booted-from-kexec");
389 }
390 
391 void smp_release_cpus(void)
392 {
393 	unsigned long *ptr;
394 	int i;
395 
396 	if (!use_spinloop())
397 		return;
398 
399 	DBG(" -> smp_release_cpus()\n");
400 
401 	/* All secondary cpus are spinning on a common spinloop, release them
402 	 * all now so they can start to spin on their individual paca
403 	 * spinloops. For non SMP kernels, the secondary cpus never get out
404 	 * of the common spinloop.
405 	 */
406 
407 	ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
408 			- PHYSICAL_START);
409 	*ptr = ppc_function_entry(generic_secondary_smp_init);
410 
411 	/* And wait a bit for them to catch up */
412 	for (i = 0; i < 100000; i++) {
413 		mb();
414 		HMT_low();
415 		if (spinning_secondaries == 0)
416 			break;
417 		udelay(1);
418 	}
419 	DBG("spinning_secondaries = %d\n", spinning_secondaries);
420 
421 	DBG(" <- smp_release_cpus()\n");
422 }
423 #endif /* CONFIG_SMP || CONFIG_KEXEC_CORE */
424 
425 /*
426  * Initialize some remaining members of the ppc64_caches and systemcfg
427  * structures
428  * (at least until we get rid of them completely). This is mostly some
429  * cache informations about the CPU that will be used by cache flush
430  * routines and/or provided to userland
431  */
432 
433 static void init_cache_info(struct ppc_cache_info *info, u32 size, u32 lsize,
434 			    u32 bsize, u32 sets)
435 {
436 	info->size = size;
437 	info->sets = sets;
438 	info->line_size = lsize;
439 	info->block_size = bsize;
440 	info->log_block_size = __ilog2(bsize);
441 	if (bsize)
442 		info->blocks_per_page = PAGE_SIZE / bsize;
443 	else
444 		info->blocks_per_page = 0;
445 
446 	if (sets == 0)
447 		info->assoc = 0xffff;
448 	else
449 		info->assoc = size / (sets * lsize);
450 }
451 
452 static bool __init parse_cache_info(struct device_node *np,
453 				    bool icache,
454 				    struct ppc_cache_info *info)
455 {
456 	static const char *ipropnames[] __initdata = {
457 		"i-cache-size",
458 		"i-cache-sets",
459 		"i-cache-block-size",
460 		"i-cache-line-size",
461 	};
462 	static const char *dpropnames[] __initdata = {
463 		"d-cache-size",
464 		"d-cache-sets",
465 		"d-cache-block-size",
466 		"d-cache-line-size",
467 	};
468 	const char **propnames = icache ? ipropnames : dpropnames;
469 	const __be32 *sizep, *lsizep, *bsizep, *setsp;
470 	u32 size, lsize, bsize, sets;
471 	bool success = true;
472 
473 	size = 0;
474 	sets = -1u;
475 	lsize = bsize = cur_cpu_spec->dcache_bsize;
476 	sizep = of_get_property(np, propnames[0], NULL);
477 	if (sizep != NULL)
478 		size = be32_to_cpu(*sizep);
479 	setsp = of_get_property(np, propnames[1], NULL);
480 	if (setsp != NULL)
481 		sets = be32_to_cpu(*setsp);
482 	bsizep = of_get_property(np, propnames[2], NULL);
483 	lsizep = of_get_property(np, propnames[3], NULL);
484 	if (bsizep == NULL)
485 		bsizep = lsizep;
486 	if (lsizep != NULL)
487 		lsize = be32_to_cpu(*lsizep);
488 	if (bsizep != NULL)
489 		bsize = be32_to_cpu(*bsizep);
490 	if (sizep == NULL || bsizep == NULL || lsizep == NULL)
491 		success = false;
492 
493 	/*
494 	 * OF is weird .. it represents fully associative caches
495 	 * as "1 way" which doesn't make much sense and doesn't
496 	 * leave room for direct mapped. We'll assume that 0
497 	 * in OF means direct mapped for that reason.
498 	 */
499 	if (sets == 1)
500 		sets = 0;
501 	else if (sets == 0)
502 		sets = 1;
503 
504 	init_cache_info(info, size, lsize, bsize, sets);
505 
506 	return success;
507 }
508 
509 void __init initialize_cache_info(void)
510 {
511 	struct device_node *cpu = NULL, *l2, *l3 = NULL;
512 	u32 pvr;
513 
514 	DBG(" -> initialize_cache_info()\n");
515 
516 	/*
517 	 * All shipping POWER8 machines have a firmware bug that
518 	 * puts incorrect information in the device-tree. This will
519 	 * be (hopefully) fixed for future chips but for now hard
520 	 * code the values if we are running on one of these
521 	 */
522 	pvr = PVR_VER(mfspr(SPRN_PVR));
523 	if (pvr == PVR_POWER8 || pvr == PVR_POWER8E ||
524 	    pvr == PVR_POWER8NVL) {
525 						/* size    lsize   blk  sets */
526 		init_cache_info(&ppc64_caches.l1i, 0x8000,   128,  128, 32);
527 		init_cache_info(&ppc64_caches.l1d, 0x10000,  128,  128, 64);
528 		init_cache_info(&ppc64_caches.l2,  0x80000,  128,  0,   512);
529 		init_cache_info(&ppc64_caches.l3,  0x800000, 128,  0,   8192);
530 	} else
531 		cpu = of_find_node_by_type(NULL, "cpu");
532 
533 	/*
534 	 * We're assuming *all* of the CPUs have the same
535 	 * d-cache and i-cache sizes... -Peter
536 	 */
537 	if (cpu) {
538 		if (!parse_cache_info(cpu, false, &ppc64_caches.l1d))
539 			DBG("Argh, can't find dcache properties !\n");
540 
541 		if (!parse_cache_info(cpu, true, &ppc64_caches.l1i))
542 			DBG("Argh, can't find icache properties !\n");
543 
544 		/*
545 		 * Try to find the L2 and L3 if any. Assume they are
546 		 * unified and use the D-side properties.
547 		 */
548 		l2 = of_find_next_cache_node(cpu);
549 		of_node_put(cpu);
550 		if (l2) {
551 			parse_cache_info(l2, false, &ppc64_caches.l2);
552 			l3 = of_find_next_cache_node(l2);
553 			of_node_put(l2);
554 		}
555 		if (l3) {
556 			parse_cache_info(l3, false, &ppc64_caches.l3);
557 			of_node_put(l3);
558 		}
559 	}
560 
561 	/* For use by binfmt_elf */
562 	dcache_bsize = ppc64_caches.l1d.block_size;
563 	icache_bsize = ppc64_caches.l1i.block_size;
564 
565 	cur_cpu_spec->dcache_bsize = dcache_bsize;
566 	cur_cpu_spec->icache_bsize = icache_bsize;
567 
568 	DBG(" <- initialize_cache_info()\n");
569 }
570 
571 /* This returns the limit below which memory accesses to the linear
572  * mapping are guarnateed not to cause a TLB or SLB miss. This is
573  * used to allocate interrupt or emergency stacks for which our
574  * exception entry path doesn't deal with being interrupted.
575  */
576 static __init u64 safe_stack_limit(void)
577 {
578 #ifdef CONFIG_PPC_BOOK3E
579 	/* Freescale BookE bolts the entire linear mapping */
580 	if (mmu_has_feature(MMU_FTR_TYPE_FSL_E))
581 		return linear_map_top;
582 	/* Other BookE, we assume the first GB is bolted */
583 	return 1ul << 30;
584 #else
585 	if (early_radix_enabled())
586 		return ULONG_MAX;
587 
588 	/* BookS, the first segment is bolted */
589 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
590 		return 1UL << SID_SHIFT_1T;
591 	return 1UL << SID_SHIFT;
592 #endif
593 }
594 
595 void __init irqstack_early_init(void)
596 {
597 	u64 limit = safe_stack_limit();
598 	unsigned int i;
599 
600 	/*
601 	 * Interrupt stacks must be in the first segment since we
602 	 * cannot afford to take SLB misses on them. They are not
603 	 * accessed in realmode.
604 	 */
605 	for_each_possible_cpu(i) {
606 		softirq_ctx[i] = (struct thread_info *)
607 			__va(memblock_alloc_base(THREAD_SIZE,
608 					    THREAD_SIZE, limit));
609 		hardirq_ctx[i] = (struct thread_info *)
610 			__va(memblock_alloc_base(THREAD_SIZE,
611 					    THREAD_SIZE, limit));
612 	}
613 }
614 
615 #ifdef CONFIG_PPC_BOOK3E
616 void __init exc_lvl_early_init(void)
617 {
618 	unsigned int i;
619 	unsigned long sp;
620 
621 	for_each_possible_cpu(i) {
622 		sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
623 		critirq_ctx[i] = (struct thread_info *)__va(sp);
624 		paca[i].crit_kstack = __va(sp + THREAD_SIZE);
625 
626 		sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
627 		dbgirq_ctx[i] = (struct thread_info *)__va(sp);
628 		paca[i].dbg_kstack = __va(sp + THREAD_SIZE);
629 
630 		sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
631 		mcheckirq_ctx[i] = (struct thread_info *)__va(sp);
632 		paca[i].mc_kstack = __va(sp + THREAD_SIZE);
633 	}
634 
635 	if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC))
636 		patch_exception(0x040, exc_debug_debug_book3e);
637 }
638 #endif
639 
640 /*
641  * Emergency stacks are used for a range of things, from asynchronous
642  * NMIs (system reset, machine check) to synchronous, process context.
643  * We set preempt_count to zero, even though that isn't necessarily correct. To
644  * get the right value we'd need to copy it from the previous thread_info, but
645  * doing that might fault causing more problems.
646  * TODO: what to do with accounting?
647  */
648 static void emerg_stack_init_thread_info(struct thread_info *ti, int cpu)
649 {
650 	ti->task = NULL;
651 	ti->cpu = cpu;
652 	ti->preempt_count = 0;
653 	ti->local_flags = 0;
654 	ti->flags = 0;
655 	klp_init_thread_info(ti);
656 }
657 
658 /*
659  * Stack space used when we detect a bad kernel stack pointer, and
660  * early in SMP boots before relocation is enabled. Exclusive emergency
661  * stack for machine checks.
662  */
663 void __init emergency_stack_init(void)
664 {
665 	u64 limit;
666 	unsigned int i;
667 
668 	/*
669 	 * Emergency stacks must be under 256MB, we cannot afford to take
670 	 * SLB misses on them. The ABI also requires them to be 128-byte
671 	 * aligned.
672 	 *
673 	 * Since we use these as temporary stacks during secondary CPU
674 	 * bringup, machine check, system reset, and HMI, we need to get
675 	 * at them in real mode. This means they must also be within the RMO
676 	 * region.
677 	 *
678 	 * The IRQ stacks allocated elsewhere in this file are zeroed and
679 	 * initialized in kernel/irq.c. These are initialized here in order
680 	 * to have emergency stacks available as early as possible.
681 	 */
682 	limit = min(safe_stack_limit(), ppc64_rma_size);
683 
684 	for_each_possible_cpu(i) {
685 		struct thread_info *ti;
686 		ti = __va(memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit));
687 		memset(ti, 0, THREAD_SIZE);
688 		emerg_stack_init_thread_info(ti, i);
689 		paca[i].emergency_sp = (void *)ti + THREAD_SIZE;
690 
691 #ifdef CONFIG_PPC_BOOK3S_64
692 		/* emergency stack for NMI exception handling. */
693 		ti = __va(memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit));
694 		memset(ti, 0, THREAD_SIZE);
695 		emerg_stack_init_thread_info(ti, i);
696 		paca[i].nmi_emergency_sp = (void *)ti + THREAD_SIZE;
697 
698 		/* emergency stack for machine check exception handling. */
699 		ti = __va(memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit));
700 		memset(ti, 0, THREAD_SIZE);
701 		emerg_stack_init_thread_info(ti, i);
702 		paca[i].mc_emergency_sp = (void *)ti + THREAD_SIZE;
703 #endif
704 	}
705 }
706 
707 #ifdef CONFIG_SMP
708 #define PCPU_DYN_SIZE		()
709 
710 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
711 {
712 	return __alloc_bootmem_node(NODE_DATA(early_cpu_to_node(cpu)), size, align,
713 				    __pa(MAX_DMA_ADDRESS));
714 }
715 
716 static void __init pcpu_fc_free(void *ptr, size_t size)
717 {
718 	free_bootmem(__pa(ptr), size);
719 }
720 
721 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
722 {
723 	if (early_cpu_to_node(from) == early_cpu_to_node(to))
724 		return LOCAL_DISTANCE;
725 	else
726 		return REMOTE_DISTANCE;
727 }
728 
729 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
730 EXPORT_SYMBOL(__per_cpu_offset);
731 
732 void __init setup_per_cpu_areas(void)
733 {
734 	const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
735 	size_t atom_size;
736 	unsigned long delta;
737 	unsigned int cpu;
738 	int rc;
739 
740 	/*
741 	 * Linear mapping is one of 4K, 1M and 16M.  For 4K, no need
742 	 * to group units.  For larger mappings, use 1M atom which
743 	 * should be large enough to contain a number of units.
744 	 */
745 	if (mmu_linear_psize == MMU_PAGE_4K)
746 		atom_size = PAGE_SIZE;
747 	else
748 		atom_size = 1 << 20;
749 
750 	rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
751 				    pcpu_fc_alloc, pcpu_fc_free);
752 	if (rc < 0)
753 		panic("cannot initialize percpu area (err=%d)", rc);
754 
755 	delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
756 	for_each_possible_cpu(cpu) {
757                 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
758 		paca[cpu].data_offset = __per_cpu_offset[cpu];
759 	}
760 }
761 #endif
762 
763 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
764 unsigned long memory_block_size_bytes(void)
765 {
766 	if (ppc_md.memory_block_size)
767 		return ppc_md.memory_block_size();
768 
769 	return MIN_MEMORY_BLOCK_SIZE;
770 }
771 #endif
772 
773 #if defined(CONFIG_PPC_INDIRECT_PIO) || defined(CONFIG_PPC_INDIRECT_MMIO)
774 struct ppc_pci_io ppc_pci_io;
775 EXPORT_SYMBOL(ppc_pci_io);
776 #endif
777 
778 #ifdef CONFIG_HARDLOCKUP_DETECTOR_PERF
779 u64 hw_nmi_get_sample_period(int watchdog_thresh)
780 {
781 	return ppc_proc_freq * watchdog_thresh;
782 }
783 #endif
784 
785 /*
786  * The perf based hardlockup detector breaks PMU event based branches, so
787  * disable it by default. Book3S has a soft-nmi hardlockup detector based
788  * on the decrementer interrupt, so it does not suffer from this problem.
789  *
790  * It is likely to get false positives in VM guests, so disable it there
791  * by default too.
792  */
793 static int __init disable_hardlockup_detector(void)
794 {
795 #ifdef CONFIG_HARDLOCKUP_DETECTOR_PERF
796 	hardlockup_detector_disable();
797 #else
798 	if (firmware_has_feature(FW_FEATURE_LPAR))
799 		hardlockup_detector_disable();
800 #endif
801 
802 	return 0;
803 }
804 early_initcall(disable_hardlockup_detector);
805 
806 #ifdef CONFIG_PPC_BOOK3S_64
807 static enum l1d_flush_type enabled_flush_types;
808 static void *l1d_flush_fallback_area;
809 static bool no_rfi_flush;
810 bool rfi_flush;
811 
812 static int __init handle_no_rfi_flush(char *p)
813 {
814 	pr_info("rfi-flush: disabled on command line.");
815 	no_rfi_flush = true;
816 	return 0;
817 }
818 early_param("no_rfi_flush", handle_no_rfi_flush);
819 
820 /*
821  * The RFI flush is not KPTI, but because users will see doco that says to use
822  * nopti we hijack that option here to also disable the RFI flush.
823  */
824 static int __init handle_no_pti(char *p)
825 {
826 	pr_info("rfi-flush: disabling due to 'nopti' on command line.\n");
827 	handle_no_rfi_flush(NULL);
828 	return 0;
829 }
830 early_param("nopti", handle_no_pti);
831 
832 static void do_nothing(void *unused)
833 {
834 	/*
835 	 * We don't need to do the flush explicitly, just enter+exit kernel is
836 	 * sufficient, the RFI exit handlers will do the right thing.
837 	 */
838 }
839 
840 void rfi_flush_enable(bool enable)
841 {
842 	if (rfi_flush == enable)
843 		return;
844 
845 	if (enable) {
846 		do_rfi_flush_fixups(enabled_flush_types);
847 		on_each_cpu(do_nothing, NULL, 1);
848 	} else
849 		do_rfi_flush_fixups(L1D_FLUSH_NONE);
850 
851 	rfi_flush = enable;
852 }
853 
854 static void init_fallback_flush(void)
855 {
856 	u64 l1d_size, limit;
857 	int cpu;
858 
859 	l1d_size = ppc64_caches.l1d.size;
860 	limit = min(safe_stack_limit(), ppc64_rma_size);
861 
862 	/*
863 	 * Align to L1d size, and size it at 2x L1d size, to catch possible
864 	 * hardware prefetch runoff. We don't have a recipe for load patterns to
865 	 * reliably avoid the prefetcher.
866 	 */
867 	l1d_flush_fallback_area = __va(memblock_alloc_base(l1d_size * 2, l1d_size, limit));
868 	memset(l1d_flush_fallback_area, 0, l1d_size * 2);
869 
870 	for_each_possible_cpu(cpu) {
871 		/*
872 		 * The fallback flush is currently coded for 8-way
873 		 * associativity. Different associativity is possible, but it
874 		 * will be treated as 8-way and may not evict the lines as
875 		 * effectively.
876 		 *
877 		 * 128 byte lines are mandatory.
878 		 */
879 		u64 c = l1d_size / 8;
880 
881 		paca[cpu].rfi_flush_fallback_area = l1d_flush_fallback_area;
882 		paca[cpu].l1d_flush_congruence = c;
883 		paca[cpu].l1d_flush_sets = c / 128;
884 	}
885 }
886 
887 void __init setup_rfi_flush(enum l1d_flush_type types, bool enable)
888 {
889 	if (types & L1D_FLUSH_FALLBACK) {
890 		pr_info("rfi-flush: Using fallback displacement flush\n");
891 		init_fallback_flush();
892 	}
893 
894 	if (types & L1D_FLUSH_ORI)
895 		pr_info("rfi-flush: Using ori type flush\n");
896 
897 	if (types & L1D_FLUSH_MTTRIG)
898 		pr_info("rfi-flush: Using mttrig type flush\n");
899 
900 	enabled_flush_types = types;
901 
902 	if (!no_rfi_flush)
903 		rfi_flush_enable(enable);
904 }
905 
906 #ifdef CONFIG_DEBUG_FS
907 static int rfi_flush_set(void *data, u64 val)
908 {
909 	if (val == 1)
910 		rfi_flush_enable(true);
911 	else if (val == 0)
912 		rfi_flush_enable(false);
913 	else
914 		return -EINVAL;
915 
916 	return 0;
917 }
918 
919 static int rfi_flush_get(void *data, u64 *val)
920 {
921 	*val = rfi_flush ? 1 : 0;
922 	return 0;
923 }
924 
925 DEFINE_SIMPLE_ATTRIBUTE(fops_rfi_flush, rfi_flush_get, rfi_flush_set, "%llu\n");
926 
927 static __init int rfi_flush_debugfs_init(void)
928 {
929 	debugfs_create_file("rfi_flush", 0600, powerpc_debugfs_root, NULL, &fops_rfi_flush);
930 	return 0;
931 }
932 device_initcall(rfi_flush_debugfs_init);
933 #endif
934 
935 ssize_t cpu_show_meltdown(struct device *dev, struct device_attribute *attr, char *buf)
936 {
937 	if (rfi_flush)
938 		return sprintf(buf, "Mitigation: RFI Flush\n");
939 
940 	return sprintf(buf, "Vulnerable\n");
941 }
942 #endif /* CONFIG_PPC_BOOK3S_64 */
943