xref: /openbmc/linux/arch/powerpc/kernel/setup_64.c (revision 9cfc5c90)
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/hugetlb.h>
39 #include <linux/memory.h>
40 #include <linux/nmi.h>
41 
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/sections.h>
54 #include <asm/btext.h>
55 #include <asm/nvram.h>
56 #include <asm/setup.h>
57 #include <asm/rtas.h>
58 #include <asm/iommu.h>
59 #include <asm/serial.h>
60 #include <asm/cache.h>
61 #include <asm/page.h>
62 #include <asm/mmu.h>
63 #include <asm/firmware.h>
64 #include <asm/xmon.h>
65 #include <asm/udbg.h>
66 #include <asm/kexec.h>
67 #include <asm/mmu_context.h>
68 #include <asm/code-patching.h>
69 #include <asm/kvm_ppc.h>
70 #include <asm/hugetlb.h>
71 #include <asm/epapr_hcalls.h>
72 
73 #ifdef DEBUG
74 #define DBG(fmt...) udbg_printf(fmt)
75 #else
76 #define DBG(fmt...)
77 #endif
78 
79 int spinning_secondaries;
80 u64 ppc64_pft_size;
81 
82 /* Pick defaults since we might want to patch instructions
83  * before we've read this from the device tree.
84  */
85 struct ppc64_caches ppc64_caches = {
86 	.dline_size = 0x40,
87 	.log_dline_size = 6,
88 	.iline_size = 0x40,
89 	.log_iline_size = 6
90 };
91 EXPORT_SYMBOL_GPL(ppc64_caches);
92 
93 /*
94  * These are used in binfmt_elf.c to put aux entries on the stack
95  * for each elf executable being started.
96  */
97 int dcache_bsize;
98 int icache_bsize;
99 int ucache_bsize;
100 
101 #if defined(CONFIG_PPC_BOOK3E) && defined(CONFIG_SMP)
102 static void setup_tlb_core_data(void)
103 {
104 	int cpu;
105 
106 	BUILD_BUG_ON(offsetof(struct tlb_core_data, lock) != 0);
107 
108 	for_each_possible_cpu(cpu) {
109 		int first = cpu_first_thread_sibling(cpu);
110 
111 		/*
112 		 * If we boot via kdump on a non-primary thread,
113 		 * make sure we point at the thread that actually
114 		 * set up this TLB.
115 		 */
116 		if (cpu_first_thread_sibling(boot_cpuid) == first)
117 			first = boot_cpuid;
118 
119 		paca[cpu].tcd_ptr = &paca[first].tcd;
120 
121 		/*
122 		 * If we have threads, we need either tlbsrx.
123 		 * or e6500 tablewalk mode, or else TLB handlers
124 		 * will be racy and could produce duplicate entries.
125 		 */
126 		if (smt_enabled_at_boot >= 2 &&
127 		    !mmu_has_feature(MMU_FTR_USE_TLBRSRV) &&
128 		    book3e_htw_mode != PPC_HTW_E6500) {
129 			/* Should we panic instead? */
130 			WARN_ONCE("%s: unsupported MMU configuration -- expect problems\n",
131 				  __func__);
132 		}
133 	}
134 }
135 #else
136 static void setup_tlb_core_data(void)
137 {
138 }
139 #endif
140 
141 #ifdef CONFIG_SMP
142 
143 static char *smt_enabled_cmdline;
144 
145 /* Look for ibm,smt-enabled OF option */
146 static void check_smt_enabled(void)
147 {
148 	struct device_node *dn;
149 	const char *smt_option;
150 
151 	/* Default to enabling all threads */
152 	smt_enabled_at_boot = threads_per_core;
153 
154 	/* Allow the command line to overrule the OF option */
155 	if (smt_enabled_cmdline) {
156 		if (!strcmp(smt_enabled_cmdline, "on"))
157 			smt_enabled_at_boot = threads_per_core;
158 		else if (!strcmp(smt_enabled_cmdline, "off"))
159 			smt_enabled_at_boot = 0;
160 		else {
161 			int smt;
162 			int rc;
163 
164 			rc = kstrtoint(smt_enabled_cmdline, 10, &smt);
165 			if (!rc)
166 				smt_enabled_at_boot =
167 					min(threads_per_core, smt);
168 		}
169 	} else {
170 		dn = of_find_node_by_path("/options");
171 		if (dn) {
172 			smt_option = of_get_property(dn, "ibm,smt-enabled",
173 						     NULL);
174 
175 			if (smt_option) {
176 				if (!strcmp(smt_option, "on"))
177 					smt_enabled_at_boot = threads_per_core;
178 				else if (!strcmp(smt_option, "off"))
179 					smt_enabled_at_boot = 0;
180 			}
181 
182 			of_node_put(dn);
183 		}
184 	}
185 }
186 
187 /* Look for smt-enabled= cmdline option */
188 static int __init early_smt_enabled(char *p)
189 {
190 	smt_enabled_cmdline = p;
191 	return 0;
192 }
193 early_param("smt-enabled", early_smt_enabled);
194 
195 #else
196 #define check_smt_enabled()
197 #endif /* CONFIG_SMP */
198 
199 /** Fix up paca fields required for the boot cpu */
200 static void fixup_boot_paca(void)
201 {
202 	/* The boot cpu is started */
203 	get_paca()->cpu_start = 1;
204 	/* Allow percpu accesses to work until we setup percpu data */
205 	get_paca()->data_offset = 0;
206 }
207 
208 static void cpu_ready_for_interrupts(void)
209 {
210 	/* Set IR and DR in PACA MSR */
211 	get_paca()->kernel_msr = MSR_KERNEL;
212 
213 	/*
214 	 * Enable AIL if supported, and we are in hypervisor mode. If we are
215 	 * not in hypervisor mode, we enable relocation-on interrupts later
216 	 * in pSeries_setup_arch() using the H_SET_MODE hcall.
217 	 */
218 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
219 	    cpu_has_feature(CPU_FTR_ARCH_207S)) {
220 		unsigned long lpcr = mfspr(SPRN_LPCR);
221 		mtspr(SPRN_LPCR, lpcr | LPCR_AIL_3);
222 	}
223 }
224 
225 /*
226  * Early initialization entry point. This is called by head.S
227  * with MMU translation disabled. We rely on the "feature" of
228  * the CPU that ignores the top 2 bits of the address in real
229  * mode so we can access kernel globals normally provided we
230  * only toy with things in the RMO region. From here, we do
231  * some early parsing of the device-tree to setup out MEMBLOCK
232  * data structures, and allocate & initialize the hash table
233  * and segment tables so we can start running with translation
234  * enabled.
235  *
236  * It is this function which will call the probe() callback of
237  * the various platform types and copy the matching one to the
238  * global ppc_md structure. Your platform can eventually do
239  * some very early initializations from the probe() routine, but
240  * this is not recommended, be very careful as, for example, the
241  * device-tree is not accessible via normal means at this point.
242  */
243 
244 void __init early_setup(unsigned long dt_ptr)
245 {
246 	static __initdata struct paca_struct boot_paca;
247 
248 	/* -------- printk is _NOT_ safe to use here ! ------- */
249 
250 	/* Identify CPU type */
251 	identify_cpu(0, mfspr(SPRN_PVR));
252 
253 	/* Assume we're on cpu 0 for now. Don't write to the paca yet! */
254 	initialise_paca(&boot_paca, 0);
255 	setup_paca(&boot_paca);
256 	fixup_boot_paca();
257 
258 	/* Initialize lockdep early or else spinlocks will blow */
259 	lockdep_init();
260 
261 	/* -------- printk is now safe to use ------- */
262 
263 	/* Enable early debugging if any specified (see udbg.h) */
264 	udbg_early_init();
265 
266  	DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
267 
268 	/*
269 	 * Do early initialization using the flattened device
270 	 * tree, such as retrieving the physical memory map or
271 	 * calculating/retrieving the hash table size.
272 	 */
273 	early_init_devtree(__va(dt_ptr));
274 
275 	epapr_paravirt_early_init();
276 
277 	/* Now we know the logical id of our boot cpu, setup the paca. */
278 	setup_paca(&paca[boot_cpuid]);
279 	fixup_boot_paca();
280 
281 	/* Probe the machine type */
282 	probe_machine();
283 
284 	setup_kdump_trampoline();
285 
286 	DBG("Found, Initializing memory management...\n");
287 
288 	/* Initialize the hash table or TLB handling */
289 	early_init_mmu();
290 
291 	/*
292 	 * At this point, we can let interrupts switch to virtual mode
293 	 * (the MMU has been setup), so adjust the MSR in the PACA to
294 	 * have IR and DR set and enable AIL if it exists
295 	 */
296 	cpu_ready_for_interrupts();
297 
298 	/* Reserve large chunks of memory for use by CMA for KVM */
299 	kvm_cma_reserve();
300 
301 	/*
302 	 * Reserve any gigantic pages requested on the command line.
303 	 * memblock needs to have been initialized by the time this is
304 	 * called since this will reserve memory.
305 	 */
306 	reserve_hugetlb_gpages();
307 
308 	DBG(" <- early_setup()\n");
309 
310 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
311 	/*
312 	 * This needs to be done *last* (after the above DBG() even)
313 	 *
314 	 * Right after we return from this function, we turn on the MMU
315 	 * which means the real-mode access trick that btext does will
316 	 * no longer work, it needs to switch to using a real MMU
317 	 * mapping. This call will ensure that it does
318 	 */
319 	btext_map();
320 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
321 }
322 
323 #ifdef CONFIG_SMP
324 void early_setup_secondary(void)
325 {
326 	/* Mark interrupts enabled in PACA */
327 	get_paca()->soft_enabled = 0;
328 
329 	/* Initialize the hash table or TLB handling */
330 	early_init_mmu_secondary();
331 
332 	/*
333 	 * At this point, we can let interrupts switch to virtual mode
334 	 * (the MMU has been setup), so adjust the MSR in the PACA to
335 	 * have IR and DR set.
336 	 */
337 	cpu_ready_for_interrupts();
338 }
339 
340 #endif /* CONFIG_SMP */
341 
342 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
343 static bool use_spinloop(void)
344 {
345 	if (!IS_ENABLED(CONFIG_PPC_BOOK3E))
346 		return true;
347 
348 	/*
349 	 * When book3e boots from kexec, the ePAPR spin table does
350 	 * not get used.
351 	 */
352 	return of_property_read_bool(of_chosen, "linux,booted-from-kexec");
353 }
354 
355 void smp_release_cpus(void)
356 {
357 	unsigned long *ptr;
358 	int i;
359 
360 	if (!use_spinloop())
361 		return;
362 
363 	DBG(" -> smp_release_cpus()\n");
364 
365 	/* All secondary cpus are spinning on a common spinloop, release them
366 	 * all now so they can start to spin on their individual paca
367 	 * spinloops. For non SMP kernels, the secondary cpus never get out
368 	 * of the common spinloop.
369 	 */
370 
371 	ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
372 			- PHYSICAL_START);
373 	*ptr = ppc_function_entry(generic_secondary_smp_init);
374 
375 	/* And wait a bit for them to catch up */
376 	for (i = 0; i < 100000; i++) {
377 		mb();
378 		HMT_low();
379 		if (spinning_secondaries == 0)
380 			break;
381 		udelay(1);
382 	}
383 	DBG("spinning_secondaries = %d\n", spinning_secondaries);
384 
385 	DBG(" <- smp_release_cpus()\n");
386 }
387 #endif /* CONFIG_SMP || CONFIG_KEXEC */
388 
389 /*
390  * Initialize some remaining members of the ppc64_caches and systemcfg
391  * structures
392  * (at least until we get rid of them completely). This is mostly some
393  * cache informations about the CPU that will be used by cache flush
394  * routines and/or provided to userland
395  */
396 static void __init initialize_cache_info(void)
397 {
398 	struct device_node *np;
399 	unsigned long num_cpus = 0;
400 
401 	DBG(" -> initialize_cache_info()\n");
402 
403 	for_each_node_by_type(np, "cpu") {
404 		num_cpus += 1;
405 
406 		/*
407 		 * We're assuming *all* of the CPUs have the same
408 		 * d-cache and i-cache sizes... -Peter
409 		 */
410 		if (num_cpus == 1) {
411 			const __be32 *sizep, *lsizep;
412 			u32 size, lsize;
413 
414 			size = 0;
415 			lsize = cur_cpu_spec->dcache_bsize;
416 			sizep = of_get_property(np, "d-cache-size", NULL);
417 			if (sizep != NULL)
418 				size = be32_to_cpu(*sizep);
419 			lsizep = of_get_property(np, "d-cache-block-size",
420 						 NULL);
421 			/* fallback if block size missing */
422 			if (lsizep == NULL)
423 				lsizep = of_get_property(np,
424 							 "d-cache-line-size",
425 							 NULL);
426 			if (lsizep != NULL)
427 				lsize = be32_to_cpu(*lsizep);
428 			if (sizep == NULL || lsizep == NULL)
429 				DBG("Argh, can't find dcache properties ! "
430 				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
431 
432 			ppc64_caches.dsize = size;
433 			ppc64_caches.dline_size = lsize;
434 			ppc64_caches.log_dline_size = __ilog2(lsize);
435 			ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
436 
437 			size = 0;
438 			lsize = cur_cpu_spec->icache_bsize;
439 			sizep = of_get_property(np, "i-cache-size", NULL);
440 			if (sizep != NULL)
441 				size = be32_to_cpu(*sizep);
442 			lsizep = of_get_property(np, "i-cache-block-size",
443 						 NULL);
444 			if (lsizep == NULL)
445 				lsizep = of_get_property(np,
446 							 "i-cache-line-size",
447 							 NULL);
448 			if (lsizep != NULL)
449 				lsize = be32_to_cpu(*lsizep);
450 			if (sizep == NULL || lsizep == NULL)
451 				DBG("Argh, can't find icache properties ! "
452 				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
453 
454 			ppc64_caches.isize = size;
455 			ppc64_caches.iline_size = lsize;
456 			ppc64_caches.log_iline_size = __ilog2(lsize);
457 			ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
458 		}
459 	}
460 
461 	DBG(" <- initialize_cache_info()\n");
462 }
463 
464 
465 /*
466  * Do some initial setup of the system.  The parameters are those which
467  * were passed in from the bootloader.
468  */
469 void __init setup_system(void)
470 {
471 	DBG(" -> setup_system()\n");
472 
473 	/* Apply the CPUs-specific and firmware specific fixups to kernel
474 	 * text (nop out sections not relevant to this CPU or this firmware)
475 	 */
476 	do_feature_fixups(cur_cpu_spec->cpu_features,
477 			  &__start___ftr_fixup, &__stop___ftr_fixup);
478 	do_feature_fixups(cur_cpu_spec->mmu_features,
479 			  &__start___mmu_ftr_fixup, &__stop___mmu_ftr_fixup);
480 	do_feature_fixups(powerpc_firmware_features,
481 			  &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
482 	do_lwsync_fixups(cur_cpu_spec->cpu_features,
483 			 &__start___lwsync_fixup, &__stop___lwsync_fixup);
484 	do_final_fixups();
485 
486 	/*
487 	 * Unflatten the device-tree passed by prom_init or kexec
488 	 */
489 	unflatten_device_tree();
490 
491 	/*
492 	 * Fill the ppc64_caches & systemcfg structures with informations
493  	 * retrieved from the device-tree.
494 	 */
495 	initialize_cache_info();
496 
497 #ifdef CONFIG_PPC_RTAS
498 	/*
499 	 * Initialize RTAS if available
500 	 */
501 	rtas_initialize();
502 #endif /* CONFIG_PPC_RTAS */
503 
504 	/*
505 	 * Check if we have an initrd provided via the device-tree
506 	 */
507 	check_for_initrd();
508 
509 	/*
510 	 * Do some platform specific early initializations, that includes
511 	 * setting up the hash table pointers. It also sets up some interrupt-mapping
512 	 * related options that will be used by finish_device_tree()
513 	 */
514 	if (ppc_md.init_early)
515 		ppc_md.init_early();
516 
517  	/*
518 	 * We can discover serial ports now since the above did setup the
519 	 * hash table management for us, thus ioremap works. We do that early
520 	 * so that further code can be debugged
521 	 */
522 	find_legacy_serial_ports();
523 
524 	/*
525 	 * Register early console
526 	 */
527 	register_early_udbg_console();
528 
529 	/*
530 	 * Initialize xmon
531 	 */
532 	xmon_setup();
533 
534 	smp_setup_cpu_maps();
535 	check_smt_enabled();
536 	setup_tlb_core_data();
537 
538 	/*
539 	 * Freescale Book3e parts spin in a loop provided by firmware,
540 	 * so smp_release_cpus() does nothing for them
541 	 */
542 #if defined(CONFIG_SMP)
543 	/* Release secondary cpus out of their spinloops at 0x60 now that
544 	 * we can map physical -> logical CPU ids
545 	 */
546 	smp_release_cpus();
547 #endif
548 
549 	pr_info("Starting Linux %s %s\n", init_utsname()->machine,
550 		 init_utsname()->version);
551 
552 	pr_info("-----------------------------------------------------\n");
553 	pr_info("ppc64_pft_size    = 0x%llx\n", ppc64_pft_size);
554 	pr_info("phys_mem_size     = 0x%llx\n", memblock_phys_mem_size());
555 
556 	if (ppc64_caches.dline_size != 0x80)
557 		pr_info("dcache_line_size  = 0x%x\n", ppc64_caches.dline_size);
558 	if (ppc64_caches.iline_size != 0x80)
559 		pr_info("icache_line_size  = 0x%x\n", ppc64_caches.iline_size);
560 
561 	pr_info("cpu_features      = 0x%016lx\n", cur_cpu_spec->cpu_features);
562 	pr_info("  possible        = 0x%016lx\n", CPU_FTRS_POSSIBLE);
563 	pr_info("  always          = 0x%016lx\n", CPU_FTRS_ALWAYS);
564 	pr_info("cpu_user_features = 0x%08x 0x%08x\n", cur_cpu_spec->cpu_user_features,
565 		cur_cpu_spec->cpu_user_features2);
566 	pr_info("mmu_features      = 0x%08x\n", cur_cpu_spec->mmu_features);
567 	pr_info("firmware_features = 0x%016lx\n", powerpc_firmware_features);
568 
569 #ifdef CONFIG_PPC_STD_MMU_64
570 	if (htab_address)
571 		pr_info("htab_address      = 0x%p\n", htab_address);
572 
573 	pr_info("htab_hash_mask    = 0x%lx\n", htab_hash_mask);
574 #endif
575 
576 	if (PHYSICAL_START > 0)
577 		pr_info("physical_start    = 0x%llx\n",
578 		       (unsigned long long)PHYSICAL_START);
579 	pr_info("-----------------------------------------------------\n");
580 
581 	DBG(" <- setup_system()\n");
582 }
583 
584 /* This returns the limit below which memory accesses to the linear
585  * mapping are guarnateed not to cause a TLB or SLB miss. This is
586  * used to allocate interrupt or emergency stacks for which our
587  * exception entry path doesn't deal with being interrupted.
588  */
589 static u64 safe_stack_limit(void)
590 {
591 #ifdef CONFIG_PPC_BOOK3E
592 	/* Freescale BookE bolts the entire linear mapping */
593 	if (mmu_has_feature(MMU_FTR_TYPE_FSL_E))
594 		return linear_map_top;
595 	/* Other BookE, we assume the first GB is bolted */
596 	return 1ul << 30;
597 #else
598 	/* BookS, the first segment is bolted */
599 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
600 		return 1UL << SID_SHIFT_1T;
601 	return 1UL << SID_SHIFT;
602 #endif
603 }
604 
605 static void __init irqstack_early_init(void)
606 {
607 	u64 limit = safe_stack_limit();
608 	unsigned int i;
609 
610 	/*
611 	 * Interrupt stacks must be in the first segment since we
612 	 * cannot afford to take SLB misses on them.
613 	 */
614 	for_each_possible_cpu(i) {
615 		softirq_ctx[i] = (struct thread_info *)
616 			__va(memblock_alloc_base(THREAD_SIZE,
617 					    THREAD_SIZE, limit));
618 		hardirq_ctx[i] = (struct thread_info *)
619 			__va(memblock_alloc_base(THREAD_SIZE,
620 					    THREAD_SIZE, limit));
621 	}
622 }
623 
624 #ifdef CONFIG_PPC_BOOK3E
625 static void __init exc_lvl_early_init(void)
626 {
627 	unsigned int i;
628 	unsigned long sp;
629 
630 	for_each_possible_cpu(i) {
631 		sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
632 		critirq_ctx[i] = (struct thread_info *)__va(sp);
633 		paca[i].crit_kstack = __va(sp + THREAD_SIZE);
634 
635 		sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
636 		dbgirq_ctx[i] = (struct thread_info *)__va(sp);
637 		paca[i].dbg_kstack = __va(sp + THREAD_SIZE);
638 
639 		sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
640 		mcheckirq_ctx[i] = (struct thread_info *)__va(sp);
641 		paca[i].mc_kstack = __va(sp + THREAD_SIZE);
642 	}
643 
644 	if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC))
645 		patch_exception(0x040, exc_debug_debug_book3e);
646 }
647 #else
648 #define exc_lvl_early_init()
649 #endif
650 
651 /*
652  * Stack space used when we detect a bad kernel stack pointer, and
653  * early in SMP boots before relocation is enabled. Exclusive emergency
654  * stack for machine checks.
655  */
656 static void __init emergency_stack_init(void)
657 {
658 	u64 limit;
659 	unsigned int i;
660 
661 	/*
662 	 * Emergency stacks must be under 256MB, we cannot afford to take
663 	 * SLB misses on them. The ABI also requires them to be 128-byte
664 	 * aligned.
665 	 *
666 	 * Since we use these as temporary stacks during secondary CPU
667 	 * bringup, we need to get at them in real mode. This means they
668 	 * must also be within the RMO region.
669 	 */
670 	limit = min(safe_stack_limit(), ppc64_rma_size);
671 
672 	for_each_possible_cpu(i) {
673 		unsigned long sp;
674 		sp  = memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
675 		sp += THREAD_SIZE;
676 		paca[i].emergency_sp = __va(sp);
677 
678 #ifdef CONFIG_PPC_BOOK3S_64
679 		/* emergency stack for machine check exception handling. */
680 		sp  = memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
681 		sp += THREAD_SIZE;
682 		paca[i].mc_emergency_sp = __va(sp);
683 #endif
684 	}
685 }
686 
687 /*
688  * Called into from start_kernel this initializes memblock, which is used
689  * to manage page allocation until mem_init is called.
690  */
691 void __init setup_arch(char **cmdline_p)
692 {
693 	*cmdline_p = boot_command_line;
694 
695 	/*
696 	 * Set cache line size based on type of cpu as a default.
697 	 * Systems with OF can look in the properties on the cpu node(s)
698 	 * for a possibly more accurate value.
699 	 */
700 	dcache_bsize = ppc64_caches.dline_size;
701 	icache_bsize = ppc64_caches.iline_size;
702 
703 	if (ppc_md.panic)
704 		setup_panic();
705 
706 	init_mm.start_code = (unsigned long)_stext;
707 	init_mm.end_code = (unsigned long) _etext;
708 	init_mm.end_data = (unsigned long) _edata;
709 	init_mm.brk = klimit;
710 #ifdef CONFIG_PPC_64K_PAGES
711 	init_mm.context.pte_frag = NULL;
712 #endif
713 #ifdef CONFIG_SPAPR_TCE_IOMMU
714 	mm_iommu_init(&init_mm.context);
715 #endif
716 	irqstack_early_init();
717 	exc_lvl_early_init();
718 	emergency_stack_init();
719 
720 	initmem_init();
721 
722 #ifdef CONFIG_DUMMY_CONSOLE
723 	conswitchp = &dummy_con;
724 #endif
725 
726 	if (ppc_md.setup_arch)
727 		ppc_md.setup_arch();
728 
729 	paging_init();
730 
731 	/* Initialize the MMU context management stuff */
732 	mmu_context_init();
733 
734 	/* Interrupt code needs to be 64K-aligned */
735 	if ((unsigned long)_stext & 0xffff)
736 		panic("Kernelbase not 64K-aligned (0x%lx)!\n",
737 		      (unsigned long)_stext);
738 }
739 
740 #ifdef CONFIG_SMP
741 #define PCPU_DYN_SIZE		()
742 
743 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
744 {
745 	return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
746 				    __pa(MAX_DMA_ADDRESS));
747 }
748 
749 static void __init pcpu_fc_free(void *ptr, size_t size)
750 {
751 	free_bootmem(__pa(ptr), size);
752 }
753 
754 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
755 {
756 	if (cpu_to_node(from) == cpu_to_node(to))
757 		return LOCAL_DISTANCE;
758 	else
759 		return REMOTE_DISTANCE;
760 }
761 
762 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
763 EXPORT_SYMBOL(__per_cpu_offset);
764 
765 void __init setup_per_cpu_areas(void)
766 {
767 	const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
768 	size_t atom_size;
769 	unsigned long delta;
770 	unsigned int cpu;
771 	int rc;
772 
773 	/*
774 	 * Linear mapping is one of 4K, 1M and 16M.  For 4K, no need
775 	 * to group units.  For larger mappings, use 1M atom which
776 	 * should be large enough to contain a number of units.
777 	 */
778 	if (mmu_linear_psize == MMU_PAGE_4K)
779 		atom_size = PAGE_SIZE;
780 	else
781 		atom_size = 1 << 20;
782 
783 	rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
784 				    pcpu_fc_alloc, pcpu_fc_free);
785 	if (rc < 0)
786 		panic("cannot initialize percpu area (err=%d)", rc);
787 
788 	delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
789 	for_each_possible_cpu(cpu) {
790                 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
791 		paca[cpu].data_offset = __per_cpu_offset[cpu];
792 	}
793 }
794 #endif
795 
796 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
797 unsigned long memory_block_size_bytes(void)
798 {
799 	if (ppc_md.memory_block_size)
800 		return ppc_md.memory_block_size();
801 
802 	return MIN_MEMORY_BLOCK_SIZE;
803 }
804 #endif
805 
806 #if defined(CONFIG_PPC_INDIRECT_PIO) || defined(CONFIG_PPC_INDIRECT_MMIO)
807 struct ppc_pci_io ppc_pci_io;
808 EXPORT_SYMBOL(ppc_pci_io);
809 #endif
810 
811 #ifdef CONFIG_HARDLOCKUP_DETECTOR
812 u64 hw_nmi_get_sample_period(int watchdog_thresh)
813 {
814 	return ppc_proc_freq * watchdog_thresh;
815 }
816 
817 /*
818  * The hardlockup detector breaks PMU event based branches and is likely
819  * to get false positives in KVM guests, so disable it by default.
820  */
821 static int __init disable_hardlockup_detector(void)
822 {
823 	hardlockup_detector_disable();
824 
825 	return 0;
826 }
827 early_initcall(disable_hardlockup_detector);
828 #endif
829