xref: /linux/arch/powerpc/kernel/setup_64.c (revision 564f7dfde24a405d877168f150ae5d29d3ad99c7)
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/io.h>
42 #include <asm/kdump.h>
43 #include <asm/prom.h>
44 #include <asm/processor.h>
45 #include <asm/pgtable.h>
46 #include <asm/smp.h>
47 #include <asm/elf.h>
48 #include <asm/machdep.h>
49 #include <asm/paca.h>
50 #include <asm/time.h>
51 #include <asm/cputable.h>
52 #include <asm/sections.h>
53 #include <asm/btext.h>
54 #include <asm/nvram.h>
55 #include <asm/setup.h>
56 #include <asm/rtas.h>
57 #include <asm/iommu.h>
58 #include <asm/serial.h>
59 #include <asm/cache.h>
60 #include <asm/page.h>
61 #include <asm/mmu.h>
62 #include <asm/firmware.h>
63 #include <asm/xmon.h>
64 #include <asm/udbg.h>
65 #include <asm/kexec.h>
66 #include <asm/code-patching.h>
67 #include <asm/livepatch.h>
68 #include <asm/opal.h>
69 #include <asm/cputhreads.h>
70 
71 #ifdef DEBUG
72 #define DBG(fmt...) udbg_printf(fmt)
73 #else
74 #define DBG(fmt...)
75 #endif
76 
77 int spinning_secondaries;
78 u64 ppc64_pft_size;
79 
80 struct ppc64_caches ppc64_caches = {
81 	.l1d = {
82 		.block_size = 0x40,
83 		.log_block_size = 6,
84 	},
85 	.l1i = {
86 		.block_size = 0x40,
87 		.log_block_size = 6
88 	},
89 };
90 EXPORT_SYMBOL_GPL(ppc64_caches);
91 
92 #if defined(CONFIG_PPC_BOOK3E) && defined(CONFIG_SMP)
93 void __init setup_tlb_core_data(void)
94 {
95 	int cpu;
96 
97 	BUILD_BUG_ON(offsetof(struct tlb_core_data, lock) != 0);
98 
99 	for_each_possible_cpu(cpu) {
100 		int first = cpu_first_thread_sibling(cpu);
101 
102 		/*
103 		 * If we boot via kdump on a non-primary thread,
104 		 * make sure we point at the thread that actually
105 		 * set up this TLB.
106 		 */
107 		if (cpu_first_thread_sibling(boot_cpuid) == first)
108 			first = boot_cpuid;
109 
110 		paca[cpu].tcd_ptr = &paca[first].tcd;
111 
112 		/*
113 		 * If we have threads, we need either tlbsrx.
114 		 * or e6500 tablewalk mode, or else TLB handlers
115 		 * will be racy and could produce duplicate entries.
116 		 * Should we panic instead?
117 		 */
118 		WARN_ONCE(smt_enabled_at_boot >= 2 &&
119 			  !mmu_has_feature(MMU_FTR_USE_TLBRSRV) &&
120 			  book3e_htw_mode != PPC_HTW_E6500,
121 			  "%s: unsupported MMU configuration\n", __func__);
122 	}
123 }
124 #endif
125 
126 #ifdef CONFIG_SMP
127 
128 static char *smt_enabled_cmdline;
129 
130 /* Look for ibm,smt-enabled OF option */
131 void __init check_smt_enabled(void)
132 {
133 	struct device_node *dn;
134 	const char *smt_option;
135 
136 	/* Default to enabling all threads */
137 	smt_enabled_at_boot = threads_per_core;
138 
139 	/* Allow the command line to overrule the OF option */
140 	if (smt_enabled_cmdline) {
141 		if (!strcmp(smt_enabled_cmdline, "on"))
142 			smt_enabled_at_boot = threads_per_core;
143 		else if (!strcmp(smt_enabled_cmdline, "off"))
144 			smt_enabled_at_boot = 0;
145 		else {
146 			int smt;
147 			int rc;
148 
149 			rc = kstrtoint(smt_enabled_cmdline, 10, &smt);
150 			if (!rc)
151 				smt_enabled_at_boot =
152 					min(threads_per_core, smt);
153 		}
154 	} else {
155 		dn = of_find_node_by_path("/options");
156 		if (dn) {
157 			smt_option = of_get_property(dn, "ibm,smt-enabled",
158 						     NULL);
159 
160 			if (smt_option) {
161 				if (!strcmp(smt_option, "on"))
162 					smt_enabled_at_boot = threads_per_core;
163 				else if (!strcmp(smt_option, "off"))
164 					smt_enabled_at_boot = 0;
165 			}
166 
167 			of_node_put(dn);
168 		}
169 	}
170 }
171 
172 /* Look for smt-enabled= cmdline option */
173 static int __init early_smt_enabled(char *p)
174 {
175 	smt_enabled_cmdline = p;
176 	return 0;
177 }
178 early_param("smt-enabled", early_smt_enabled);
179 
180 #endif /* CONFIG_SMP */
181 
182 /** Fix up paca fields required for the boot cpu */
183 static void __init fixup_boot_paca(void)
184 {
185 	/* The boot cpu is started */
186 	get_paca()->cpu_start = 1;
187 	/* Allow percpu accesses to work until we setup percpu data */
188 	get_paca()->data_offset = 0;
189 }
190 
191 static void __init configure_exceptions(void)
192 {
193 	/*
194 	 * Setup the trampolines from the lowmem exception vectors
195 	 * to the kdump kernel when not using a relocatable kernel.
196 	 */
197 	setup_kdump_trampoline();
198 
199 	/* Under a PAPR hypervisor, we need hypercalls */
200 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
201 		/* Enable AIL if possible */
202 		pseries_enable_reloc_on_exc();
203 
204 		/*
205 		 * Tell the hypervisor that we want our exceptions to
206 		 * be taken in little endian mode.
207 		 *
208 		 * We don't call this for big endian as our calling convention
209 		 * makes us always enter in BE, and the call may fail under
210 		 * some circumstances with kdump.
211 		 */
212 #ifdef __LITTLE_ENDIAN__
213 		pseries_little_endian_exceptions();
214 #endif
215 	} else {
216 		/* Set endian mode using OPAL */
217 		if (firmware_has_feature(FW_FEATURE_OPAL))
218 			opal_configure_cores();
219 
220 		/* AIL on native is done in cpu_ready_for_interrupts() */
221 	}
222 }
223 
224 static void cpu_ready_for_interrupts(void)
225 {
226 	/*
227 	 * Enable AIL if supported, and we are in hypervisor mode. This
228 	 * is called once for every processor.
229 	 *
230 	 * If we are not in hypervisor mode the job is done once for
231 	 * the whole partition in configure_exceptions().
232 	 */
233 	if (early_cpu_has_feature(CPU_FTR_HVMODE) &&
234 	    early_cpu_has_feature(CPU_FTR_ARCH_207S)) {
235 		unsigned long lpcr = mfspr(SPRN_LPCR);
236 		mtspr(SPRN_LPCR, lpcr | LPCR_AIL_3);
237 	}
238 
239 	/*
240 	 * Fixup HFSCR:TM based on CPU features. The bit is set by our
241 	 * early asm init because at that point we haven't updated our
242 	 * CPU features from firmware and device-tree. Here we have,
243 	 * so let's do it.
244 	 */
245 	if (cpu_has_feature(CPU_FTR_HVMODE) && !cpu_has_feature(CPU_FTR_TM_COMP))
246 		mtspr(SPRN_HFSCR, mfspr(SPRN_HFSCR) & ~HFSCR_TM);
247 
248 	/* Set IR and DR in PACA MSR */
249 	get_paca()->kernel_msr = MSR_KERNEL;
250 }
251 
252 /*
253  * Early initialization entry point. This is called by head.S
254  * with MMU translation disabled. We rely on the "feature" of
255  * the CPU that ignores the top 2 bits of the address in real
256  * mode so we can access kernel globals normally provided we
257  * only toy with things in the RMO region. From here, we do
258  * some early parsing of the device-tree to setup out MEMBLOCK
259  * data structures, and allocate & initialize the hash table
260  * and segment tables so we can start running with translation
261  * enabled.
262  *
263  * It is this function which will call the probe() callback of
264  * the various platform types and copy the matching one to the
265  * global ppc_md structure. Your platform can eventually do
266  * some very early initializations from the probe() routine, but
267  * this is not recommended, be very careful as, for example, the
268  * device-tree is not accessible via normal means at this point.
269  */
270 
271 void __init early_setup(unsigned long dt_ptr)
272 {
273 	static __initdata struct paca_struct boot_paca;
274 
275 	/* -------- printk is _NOT_ safe to use here ! ------- */
276 
277 	/* Identify CPU type */
278 	identify_cpu(0, mfspr(SPRN_PVR));
279 
280 	/* Assume we're on cpu 0 for now. Don't write to the paca yet! */
281 	initialise_paca(&boot_paca, 0);
282 	setup_paca(&boot_paca);
283 	fixup_boot_paca();
284 
285 	/* -------- printk is now safe to use ------- */
286 
287 	/* Enable early debugging if any specified (see udbg.h) */
288 	udbg_early_init();
289 
290  	DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
291 
292 	/*
293 	 * Do early initialization using the flattened device
294 	 * tree, such as retrieving the physical memory map or
295 	 * calculating/retrieving the hash table size.
296 	 */
297 	early_init_devtree(__va(dt_ptr));
298 
299 	/* Now we know the logical id of our boot cpu, setup the paca. */
300 	setup_paca(&paca[boot_cpuid]);
301 	fixup_boot_paca();
302 
303 	/*
304 	 * Configure exception handlers. This include setting up trampolines
305 	 * if needed, setting exception endian mode, etc...
306 	 */
307 	configure_exceptions();
308 
309 	/* Apply all the dynamic patching */
310 	apply_feature_fixups();
311 	setup_feature_keys();
312 
313 	/* Initialize the hash table or TLB handling */
314 	early_init_mmu();
315 
316 	/*
317 	 * At this point, we can let interrupts switch to virtual mode
318 	 * (the MMU has been setup), so adjust the MSR in the PACA to
319 	 * have IR and DR set and enable AIL if it exists
320 	 */
321 	cpu_ready_for_interrupts();
322 
323 	DBG(" <- early_setup()\n");
324 
325 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
326 	/*
327 	 * This needs to be done *last* (after the above DBG() even)
328 	 *
329 	 * Right after we return from this function, we turn on the MMU
330 	 * which means the real-mode access trick that btext does will
331 	 * no longer work, it needs to switch to using a real MMU
332 	 * mapping. This call will ensure that it does
333 	 */
334 	btext_map();
335 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
336 }
337 
338 #ifdef CONFIG_SMP
339 void early_setup_secondary(void)
340 {
341 	/* Mark interrupts disabled in PACA */
342 	get_paca()->soft_enabled = 0;
343 
344 	/* Initialize the hash table or TLB handling */
345 	early_init_mmu_secondary();
346 
347 	/*
348 	 * At this point, we can let interrupts switch to virtual mode
349 	 * (the MMU has been setup), so adjust the MSR in the PACA to
350 	 * have IR and DR set.
351 	 */
352 	cpu_ready_for_interrupts();
353 }
354 
355 #endif /* CONFIG_SMP */
356 
357 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC_CORE)
358 static bool use_spinloop(void)
359 {
360 	if (!IS_ENABLED(CONFIG_PPC_BOOK3E))
361 		return true;
362 
363 	/*
364 	 * When book3e boots from kexec, the ePAPR spin table does
365 	 * not get used.
366 	 */
367 	return of_property_read_bool(of_chosen, "linux,booted-from-kexec");
368 }
369 
370 void smp_release_cpus(void)
371 {
372 	unsigned long *ptr;
373 	int i;
374 
375 	if (!use_spinloop())
376 		return;
377 
378 	DBG(" -> smp_release_cpus()\n");
379 
380 	/* All secondary cpus are spinning on a common spinloop, release them
381 	 * all now so they can start to spin on their individual paca
382 	 * spinloops. For non SMP kernels, the secondary cpus never get out
383 	 * of the common spinloop.
384 	 */
385 
386 	ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
387 			- PHYSICAL_START);
388 	*ptr = ppc_function_entry(generic_secondary_smp_init);
389 
390 	/* And wait a bit for them to catch up */
391 	for (i = 0; i < 100000; i++) {
392 		mb();
393 		HMT_low();
394 		if (spinning_secondaries == 0)
395 			break;
396 		udelay(1);
397 	}
398 	DBG("spinning_secondaries = %d\n", spinning_secondaries);
399 
400 	DBG(" <- smp_release_cpus()\n");
401 }
402 #endif /* CONFIG_SMP || CONFIG_KEXEC_CORE */
403 
404 /*
405  * Initialize some remaining members of the ppc64_caches and systemcfg
406  * structures
407  * (at least until we get rid of them completely). This is mostly some
408  * cache informations about the CPU that will be used by cache flush
409  * routines and/or provided to userland
410  */
411 
412 static void init_cache_info(struct ppc_cache_info *info, u32 size, u32 lsize,
413 			    u32 bsize, u32 sets)
414 {
415 	info->size = size;
416 	info->sets = sets;
417 	info->line_size = lsize;
418 	info->block_size = bsize;
419 	info->log_block_size = __ilog2(bsize);
420 	if (bsize)
421 		info->blocks_per_page = PAGE_SIZE / bsize;
422 	else
423 		info->blocks_per_page = 0;
424 
425 	if (sets == 0)
426 		info->assoc = 0xffff;
427 	else
428 		info->assoc = size / (sets * lsize);
429 }
430 
431 static bool __init parse_cache_info(struct device_node *np,
432 				    bool icache,
433 				    struct ppc_cache_info *info)
434 {
435 	static const char *ipropnames[] __initdata = {
436 		"i-cache-size",
437 		"i-cache-sets",
438 		"i-cache-block-size",
439 		"i-cache-line-size",
440 	};
441 	static const char *dpropnames[] __initdata = {
442 		"d-cache-size",
443 		"d-cache-sets",
444 		"d-cache-block-size",
445 		"d-cache-line-size",
446 	};
447 	const char **propnames = icache ? ipropnames : dpropnames;
448 	const __be32 *sizep, *lsizep, *bsizep, *setsp;
449 	u32 size, lsize, bsize, sets;
450 	bool success = true;
451 
452 	size = 0;
453 	sets = -1u;
454 	lsize = bsize = cur_cpu_spec->dcache_bsize;
455 	sizep = of_get_property(np, propnames[0], NULL);
456 	if (sizep != NULL)
457 		size = be32_to_cpu(*sizep);
458 	setsp = of_get_property(np, propnames[1], NULL);
459 	if (setsp != NULL)
460 		sets = be32_to_cpu(*setsp);
461 	bsizep = of_get_property(np, propnames[2], NULL);
462 	lsizep = of_get_property(np, propnames[3], NULL);
463 	if (bsizep == NULL)
464 		bsizep = lsizep;
465 	if (lsizep != NULL)
466 		lsize = be32_to_cpu(*lsizep);
467 	if (bsizep != NULL)
468 		bsize = be32_to_cpu(*bsizep);
469 	if (sizep == NULL || bsizep == NULL || lsizep == NULL)
470 		success = false;
471 
472 	/*
473 	 * OF is weird .. it represents fully associative caches
474 	 * as "1 way" which doesn't make much sense and doesn't
475 	 * leave room for direct mapped. We'll assume that 0
476 	 * in OF means direct mapped for that reason.
477 	 */
478 	if (sets == 1)
479 		sets = 0;
480 	else if (sets == 0)
481 		sets = 1;
482 
483 	init_cache_info(info, size, lsize, bsize, sets);
484 
485 	return success;
486 }
487 
488 void __init initialize_cache_info(void)
489 {
490 	struct device_node *cpu = NULL, *l2, *l3 = NULL;
491 	u32 pvr;
492 
493 	DBG(" -> initialize_cache_info()\n");
494 
495 	/*
496 	 * All shipping POWER8 machines have a firmware bug that
497 	 * puts incorrect information in the device-tree. This will
498 	 * be (hopefully) fixed for future chips but for now hard
499 	 * code the values if we are running on one of these
500 	 */
501 	pvr = PVR_VER(mfspr(SPRN_PVR));
502 	if (pvr == PVR_POWER8 || pvr == PVR_POWER8E ||
503 	    pvr == PVR_POWER8NVL) {
504 						/* size    lsize   blk  sets */
505 		init_cache_info(&ppc64_caches.l1i, 0x8000,   128,  128, 32);
506 		init_cache_info(&ppc64_caches.l1d, 0x10000,  128,  128, 64);
507 		init_cache_info(&ppc64_caches.l2,  0x80000,  128,  0,   512);
508 		init_cache_info(&ppc64_caches.l3,  0x800000, 128,  0,   8192);
509 	} else
510 		cpu = of_find_node_by_type(NULL, "cpu");
511 
512 	/*
513 	 * We're assuming *all* of the CPUs have the same
514 	 * d-cache and i-cache sizes... -Peter
515 	 */
516 	if (cpu) {
517 		if (!parse_cache_info(cpu, false, &ppc64_caches.l1d))
518 			DBG("Argh, can't find dcache properties !\n");
519 
520 		if (!parse_cache_info(cpu, true, &ppc64_caches.l1i))
521 			DBG("Argh, can't find icache properties !\n");
522 
523 		/*
524 		 * Try to find the L2 and L3 if any. Assume they are
525 		 * unified and use the D-side properties.
526 		 */
527 		l2 = of_find_next_cache_node(cpu);
528 		of_node_put(cpu);
529 		if (l2) {
530 			parse_cache_info(l2, false, &ppc64_caches.l2);
531 			l3 = of_find_next_cache_node(l2);
532 			of_node_put(l2);
533 		}
534 		if (l3) {
535 			parse_cache_info(l3, false, &ppc64_caches.l3);
536 			of_node_put(l3);
537 		}
538 	}
539 
540 	/* For use by binfmt_elf */
541 	dcache_bsize = ppc64_caches.l1d.block_size;
542 	icache_bsize = ppc64_caches.l1i.block_size;
543 
544 	DBG(" <- initialize_cache_info()\n");
545 }
546 
547 /* This returns the limit below which memory accesses to the linear
548  * mapping are guarnateed not to cause a TLB or SLB miss. This is
549  * used to allocate interrupt or emergency stacks for which our
550  * exception entry path doesn't deal with being interrupted.
551  */
552 static __init u64 safe_stack_limit(void)
553 {
554 #ifdef CONFIG_PPC_BOOK3E
555 	/* Freescale BookE bolts the entire linear mapping */
556 	if (mmu_has_feature(MMU_FTR_TYPE_FSL_E))
557 		return linear_map_top;
558 	/* Other BookE, we assume the first GB is bolted */
559 	return 1ul << 30;
560 #else
561 	/* BookS, the first segment is bolted */
562 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
563 		return 1UL << SID_SHIFT_1T;
564 	return 1UL << SID_SHIFT;
565 #endif
566 }
567 
568 void __init irqstack_early_init(void)
569 {
570 	u64 limit = safe_stack_limit();
571 	unsigned int i;
572 
573 	/*
574 	 * Interrupt stacks must be in the first segment since we
575 	 * cannot afford to take SLB misses on them.
576 	 */
577 	for_each_possible_cpu(i) {
578 		softirq_ctx[i] = (struct thread_info *)
579 			__va(memblock_alloc_base(THREAD_SIZE,
580 					    THREAD_SIZE, limit));
581 		hardirq_ctx[i] = (struct thread_info *)
582 			__va(memblock_alloc_base(THREAD_SIZE,
583 					    THREAD_SIZE, limit));
584 	}
585 }
586 
587 #ifdef CONFIG_PPC_BOOK3E
588 void __init exc_lvl_early_init(void)
589 {
590 	unsigned int i;
591 	unsigned long sp;
592 
593 	for_each_possible_cpu(i) {
594 		sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
595 		critirq_ctx[i] = (struct thread_info *)__va(sp);
596 		paca[i].crit_kstack = __va(sp + THREAD_SIZE);
597 
598 		sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
599 		dbgirq_ctx[i] = (struct thread_info *)__va(sp);
600 		paca[i].dbg_kstack = __va(sp + THREAD_SIZE);
601 
602 		sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
603 		mcheckirq_ctx[i] = (struct thread_info *)__va(sp);
604 		paca[i].mc_kstack = __va(sp + THREAD_SIZE);
605 	}
606 
607 	if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC))
608 		patch_exception(0x040, exc_debug_debug_book3e);
609 }
610 #endif
611 
612 /*
613  * Stack space used when we detect a bad kernel stack pointer, and
614  * early in SMP boots before relocation is enabled. Exclusive emergency
615  * stack for machine checks.
616  */
617 void __init emergency_stack_init(void)
618 {
619 	u64 limit;
620 	unsigned int i;
621 
622 	/*
623 	 * Emergency stacks must be under 256MB, we cannot afford to take
624 	 * SLB misses on them. The ABI also requires them to be 128-byte
625 	 * aligned.
626 	 *
627 	 * Since we use these as temporary stacks during secondary CPU
628 	 * bringup, we need to get at them in real mode. This means they
629 	 * must also be within the RMO region.
630 	 */
631 	limit = min(safe_stack_limit(), ppc64_rma_size);
632 
633 	for_each_possible_cpu(i) {
634 		struct thread_info *ti;
635 		ti = __va(memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit));
636 		klp_init_thread_info(ti);
637 		paca[i].emergency_sp = (void *)ti + THREAD_SIZE;
638 
639 #ifdef CONFIG_PPC_BOOK3S_64
640 		/* emergency stack for machine check exception handling. */
641 		ti = __va(memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit));
642 		klp_init_thread_info(ti);
643 		paca[i].mc_emergency_sp = (void *)ti + THREAD_SIZE;
644 #endif
645 	}
646 }
647 
648 #ifdef CONFIG_SMP
649 #define PCPU_DYN_SIZE		()
650 
651 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
652 {
653 	return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
654 				    __pa(MAX_DMA_ADDRESS));
655 }
656 
657 static void __init pcpu_fc_free(void *ptr, size_t size)
658 {
659 	free_bootmem(__pa(ptr), size);
660 }
661 
662 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
663 {
664 	if (cpu_to_node(from) == cpu_to_node(to))
665 		return LOCAL_DISTANCE;
666 	else
667 		return REMOTE_DISTANCE;
668 }
669 
670 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
671 EXPORT_SYMBOL(__per_cpu_offset);
672 
673 void __init setup_per_cpu_areas(void)
674 {
675 	const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
676 	size_t atom_size;
677 	unsigned long delta;
678 	unsigned int cpu;
679 	int rc;
680 
681 	/*
682 	 * Linear mapping is one of 4K, 1M and 16M.  For 4K, no need
683 	 * to group units.  For larger mappings, use 1M atom which
684 	 * should be large enough to contain a number of units.
685 	 */
686 	if (mmu_linear_psize == MMU_PAGE_4K)
687 		atom_size = PAGE_SIZE;
688 	else
689 		atom_size = 1 << 20;
690 
691 	rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
692 				    pcpu_fc_alloc, pcpu_fc_free);
693 	if (rc < 0)
694 		panic("cannot initialize percpu area (err=%d)", rc);
695 
696 	delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
697 	for_each_possible_cpu(cpu) {
698                 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
699 		paca[cpu].data_offset = __per_cpu_offset[cpu];
700 	}
701 }
702 #endif
703 
704 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
705 unsigned long memory_block_size_bytes(void)
706 {
707 	if (ppc_md.memory_block_size)
708 		return ppc_md.memory_block_size();
709 
710 	return MIN_MEMORY_BLOCK_SIZE;
711 }
712 #endif
713 
714 #if defined(CONFIG_PPC_INDIRECT_PIO) || defined(CONFIG_PPC_INDIRECT_MMIO)
715 struct ppc_pci_io ppc_pci_io;
716 EXPORT_SYMBOL(ppc_pci_io);
717 #endif
718 
719 #ifdef CONFIG_HARDLOCKUP_DETECTOR
720 u64 hw_nmi_get_sample_period(int watchdog_thresh)
721 {
722 	return ppc_proc_freq * watchdog_thresh;
723 }
724 
725 /*
726  * The hardlockup detector breaks PMU event based branches and is likely
727  * to get false positives in KVM guests, so disable it by default.
728  */
729 static int __init disable_hardlockup_detector(void)
730 {
731 	hardlockup_detector_disable();
732 
733 	return 0;
734 }
735 early_initcall(disable_hardlockup_detector);
736 #endif
737