xref: /linux/arch/powerpc/kernel/setup_64.c (revision 071bf69a0220253a44acb8b2a27f7a262b9a46bf)
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 /* Pick defaults since we might want to patch instructions
81  * before we've read this from the device tree.
82  */
83 struct ppc64_caches ppc64_caches = {
84 	.dline_size = 0x40,
85 	.log_dline_size = 6,
86 	.iline_size = 0x40,
87 	.log_iline_size = 6
88 };
89 EXPORT_SYMBOL_GPL(ppc64_caches);
90 
91 /*
92  * These are used in binfmt_elf.c to put aux entries on the stack
93  * for each elf executable being started.
94  */
95 int dcache_bsize;
96 int icache_bsize;
97 int ucache_bsize;
98 
99 #if defined(CONFIG_PPC_BOOK3E) && defined(CONFIG_SMP)
100 void __init setup_tlb_core_data(void)
101 {
102 	int cpu;
103 
104 	BUILD_BUG_ON(offsetof(struct tlb_core_data, lock) != 0);
105 
106 	for_each_possible_cpu(cpu) {
107 		int first = cpu_first_thread_sibling(cpu);
108 
109 		/*
110 		 * If we boot via kdump on a non-primary thread,
111 		 * make sure we point at the thread that actually
112 		 * set up this TLB.
113 		 */
114 		if (cpu_first_thread_sibling(boot_cpuid) == first)
115 			first = boot_cpuid;
116 
117 		paca[cpu].tcd_ptr = &paca[first].tcd;
118 
119 		/*
120 		 * If we have threads, we need either tlbsrx.
121 		 * or e6500 tablewalk mode, or else TLB handlers
122 		 * will be racy and could produce duplicate entries.
123 		 */
124 		if (smt_enabled_at_boot >= 2 &&
125 		    !mmu_has_feature(MMU_FTR_USE_TLBRSRV) &&
126 		    book3e_htw_mode != PPC_HTW_E6500) {
127 			/* Should we panic instead? */
128 			WARN_ONCE("%s: unsupported MMU configuration -- expect problems\n",
129 				  __func__);
130 		}
131 	}
132 }
133 #endif
134 
135 #ifdef CONFIG_SMP
136 
137 static char *smt_enabled_cmdline;
138 
139 /* Look for ibm,smt-enabled OF option */
140 void __init check_smt_enabled(void)
141 {
142 	struct device_node *dn;
143 	const char *smt_option;
144 
145 	/* Default to enabling all threads */
146 	smt_enabled_at_boot = threads_per_core;
147 
148 	/* Allow the command line to overrule the OF option */
149 	if (smt_enabled_cmdline) {
150 		if (!strcmp(smt_enabled_cmdline, "on"))
151 			smt_enabled_at_boot = threads_per_core;
152 		else if (!strcmp(smt_enabled_cmdline, "off"))
153 			smt_enabled_at_boot = 0;
154 		else {
155 			int smt;
156 			int rc;
157 
158 			rc = kstrtoint(smt_enabled_cmdline, 10, &smt);
159 			if (!rc)
160 				smt_enabled_at_boot =
161 					min(threads_per_core, smt);
162 		}
163 	} else {
164 		dn = of_find_node_by_path("/options");
165 		if (dn) {
166 			smt_option = of_get_property(dn, "ibm,smt-enabled",
167 						     NULL);
168 
169 			if (smt_option) {
170 				if (!strcmp(smt_option, "on"))
171 					smt_enabled_at_boot = threads_per_core;
172 				else if (!strcmp(smt_option, "off"))
173 					smt_enabled_at_boot = 0;
174 			}
175 
176 			of_node_put(dn);
177 		}
178 	}
179 }
180 
181 /* Look for smt-enabled= cmdline option */
182 static int __init early_smt_enabled(char *p)
183 {
184 	smt_enabled_cmdline = p;
185 	return 0;
186 }
187 early_param("smt-enabled", early_smt_enabled);
188 
189 #endif /* CONFIG_SMP */
190 
191 /** Fix up paca fields required for the boot cpu */
192 static void __init fixup_boot_paca(void)
193 {
194 	/* The boot cpu is started */
195 	get_paca()->cpu_start = 1;
196 	/* Allow percpu accesses to work until we setup percpu data */
197 	get_paca()->data_offset = 0;
198 }
199 
200 static void __init configure_exceptions(void)
201 {
202 	/*
203 	 * Setup the trampolines from the lowmem exception vectors
204 	 * to the kdump kernel when not using a relocatable kernel.
205 	 */
206 	setup_kdump_trampoline();
207 
208 	/* Under a PAPR hypervisor, we need hypercalls */
209 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
210 		/* Enable AIL if possible */
211 		pseries_enable_reloc_on_exc();
212 
213 		/*
214 		 * Tell the hypervisor that we want our exceptions to
215 		 * be taken in little endian mode.
216 		 *
217 		 * We don't call this for big endian as our calling convention
218 		 * makes us always enter in BE, and the call may fail under
219 		 * some circumstances with kdump.
220 		 */
221 #ifdef __LITTLE_ENDIAN__
222 		pseries_little_endian_exceptions();
223 #endif
224 	} else {
225 		/* Set endian mode using OPAL */
226 		if (firmware_has_feature(FW_FEATURE_OPAL))
227 			opal_configure_cores();
228 
229 		/* Enable AIL if supported, and we are in hypervisor mode */
230 		if (early_cpu_has_feature(CPU_FTR_HVMODE) &&
231 		    early_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 
238 static void cpu_ready_for_interrupts(void)
239 {
240 	/* Set IR and DR in PACA MSR */
241 	get_paca()->kernel_msr = MSR_KERNEL;
242 }
243 
244 /*
245  * Early initialization entry point. This is called by head.S
246  * with MMU translation disabled. We rely on the "feature" of
247  * the CPU that ignores the top 2 bits of the address in real
248  * mode so we can access kernel globals normally provided we
249  * only toy with things in the RMO region. From here, we do
250  * some early parsing of the device-tree to setup out MEMBLOCK
251  * data structures, and allocate & initialize the hash table
252  * and segment tables so we can start running with translation
253  * enabled.
254  *
255  * It is this function which will call the probe() callback of
256  * the various platform types and copy the matching one to the
257  * global ppc_md structure. Your platform can eventually do
258  * some very early initializations from the probe() routine, but
259  * this is not recommended, be very careful as, for example, the
260  * device-tree is not accessible via normal means at this point.
261  */
262 
263 void __init early_setup(unsigned long dt_ptr)
264 {
265 	static __initdata struct paca_struct boot_paca;
266 
267 	/* -------- printk is _NOT_ safe to use here ! ------- */
268 
269 	/* Identify CPU type */
270 	identify_cpu(0, mfspr(SPRN_PVR));
271 
272 	/* Assume we're on cpu 0 for now. Don't write to the paca yet! */
273 	initialise_paca(&boot_paca, 0);
274 	setup_paca(&boot_paca);
275 	fixup_boot_paca();
276 
277 	/* -------- printk is now safe to use ------- */
278 
279 	/* Enable early debugging if any specified (see udbg.h) */
280 	udbg_early_init();
281 
282  	DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
283 
284 	/*
285 	 * Do early initialization using the flattened device
286 	 * tree, such as retrieving the physical memory map or
287 	 * calculating/retrieving the hash table size.
288 	 */
289 	early_init_devtree(__va(dt_ptr));
290 
291 	/* Now we know the logical id of our boot cpu, setup the paca. */
292 	setup_paca(&paca[boot_cpuid]);
293 	fixup_boot_paca();
294 
295 	/*
296 	 * Configure exception handlers. This include setting up trampolines
297 	 * if needed, setting exception endian mode, etc...
298 	 */
299 	configure_exceptions();
300 
301 	/* Apply all the dynamic patching */
302 	apply_feature_fixups();
303 
304 	/* Initialize the hash table or TLB handling */
305 	early_init_mmu();
306 
307 	/*
308 	 * At this point, we can let interrupts switch to virtual mode
309 	 * (the MMU has been setup), so adjust the MSR in the PACA to
310 	 * have IR and DR set and enable AIL if it exists
311 	 */
312 	cpu_ready_for_interrupts();
313 
314 	DBG(" <- early_setup()\n");
315 
316 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
317 	/*
318 	 * This needs to be done *last* (after the above DBG() even)
319 	 *
320 	 * Right after we return from this function, we turn on the MMU
321 	 * which means the real-mode access trick that btext does will
322 	 * no longer work, it needs to switch to using a real MMU
323 	 * mapping. This call will ensure that it does
324 	 */
325 	btext_map();
326 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
327 }
328 
329 #ifdef CONFIG_SMP
330 void early_setup_secondary(void)
331 {
332 	/* Mark interrupts disabled in PACA */
333 	get_paca()->soft_enabled = 0;
334 
335 	/* Initialize the hash table or TLB handling */
336 	early_init_mmu_secondary();
337 
338 	/*
339 	 * At this point, we can let interrupts switch to virtual mode
340 	 * (the MMU has been setup), so adjust the MSR in the PACA to
341 	 * have IR and DR set.
342 	 */
343 	cpu_ready_for_interrupts();
344 }
345 
346 #endif /* CONFIG_SMP */
347 
348 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
349 static bool use_spinloop(void)
350 {
351 	if (!IS_ENABLED(CONFIG_PPC_BOOK3E))
352 		return true;
353 
354 	/*
355 	 * When book3e boots from kexec, the ePAPR spin table does
356 	 * not get used.
357 	 */
358 	return of_property_read_bool(of_chosen, "linux,booted-from-kexec");
359 }
360 
361 void smp_release_cpus(void)
362 {
363 	unsigned long *ptr;
364 	int i;
365 
366 	if (!use_spinloop())
367 		return;
368 
369 	DBG(" -> smp_release_cpus()\n");
370 
371 	/* All secondary cpus are spinning on a common spinloop, release them
372 	 * all now so they can start to spin on their individual paca
373 	 * spinloops. For non SMP kernels, the secondary cpus never get out
374 	 * of the common spinloop.
375 	 */
376 
377 	ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
378 			- PHYSICAL_START);
379 	*ptr = ppc_function_entry(generic_secondary_smp_init);
380 
381 	/* And wait a bit for them to catch up */
382 	for (i = 0; i < 100000; i++) {
383 		mb();
384 		HMT_low();
385 		if (spinning_secondaries == 0)
386 			break;
387 		udelay(1);
388 	}
389 	DBG("spinning_secondaries = %d\n", spinning_secondaries);
390 
391 	DBG(" <- smp_release_cpus()\n");
392 }
393 #endif /* CONFIG_SMP || CONFIG_KEXEC */
394 
395 /*
396  * Initialize some remaining members of the ppc64_caches and systemcfg
397  * structures
398  * (at least until we get rid of them completely). This is mostly some
399  * cache informations about the CPU that will be used by cache flush
400  * routines and/or provided to userland
401  */
402 void __init initialize_cache_info(void)
403 {
404 	struct device_node *np;
405 	unsigned long num_cpus = 0;
406 
407 	DBG(" -> initialize_cache_info()\n");
408 
409 	for_each_node_by_type(np, "cpu") {
410 		num_cpus += 1;
411 
412 		/*
413 		 * We're assuming *all* of the CPUs have the same
414 		 * d-cache and i-cache sizes... -Peter
415 		 */
416 		if (num_cpus == 1) {
417 			const __be32 *sizep, *lsizep;
418 			u32 size, lsize;
419 
420 			size = 0;
421 			lsize = cur_cpu_spec->dcache_bsize;
422 			sizep = of_get_property(np, "d-cache-size", NULL);
423 			if (sizep != NULL)
424 				size = be32_to_cpu(*sizep);
425 			lsizep = of_get_property(np, "d-cache-block-size",
426 						 NULL);
427 			/* fallback if block size missing */
428 			if (lsizep == NULL)
429 				lsizep = of_get_property(np,
430 							 "d-cache-line-size",
431 							 NULL);
432 			if (lsizep != NULL)
433 				lsize = be32_to_cpu(*lsizep);
434 			if (sizep == NULL || lsizep == NULL)
435 				DBG("Argh, can't find dcache properties ! "
436 				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
437 
438 			ppc64_caches.dsize = size;
439 			ppc64_caches.dline_size = lsize;
440 			ppc64_caches.log_dline_size = __ilog2(lsize);
441 			ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
442 
443 			size = 0;
444 			lsize = cur_cpu_spec->icache_bsize;
445 			sizep = of_get_property(np, "i-cache-size", NULL);
446 			if (sizep != NULL)
447 				size = be32_to_cpu(*sizep);
448 			lsizep = of_get_property(np, "i-cache-block-size",
449 						 NULL);
450 			if (lsizep == NULL)
451 				lsizep = of_get_property(np,
452 							 "i-cache-line-size",
453 							 NULL);
454 			if (lsizep != NULL)
455 				lsize = be32_to_cpu(*lsizep);
456 			if (sizep == NULL || lsizep == NULL)
457 				DBG("Argh, can't find icache properties ! "
458 				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
459 
460 			ppc64_caches.isize = size;
461 			ppc64_caches.iline_size = lsize;
462 			ppc64_caches.log_iline_size = __ilog2(lsize);
463 			ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
464 		}
465 	}
466 
467 	/* For use by binfmt_elf */
468 	dcache_bsize = ppc64_caches.dline_size;
469 	icache_bsize = ppc64_caches.iline_size;
470 
471 	DBG(" <- initialize_cache_info()\n");
472 }
473 
474 /* This returns the limit below which memory accesses to the linear
475  * mapping are guarnateed not to cause a TLB or SLB miss. This is
476  * used to allocate interrupt or emergency stacks for which our
477  * exception entry path doesn't deal with being interrupted.
478  */
479 static __init u64 safe_stack_limit(void)
480 {
481 #ifdef CONFIG_PPC_BOOK3E
482 	/* Freescale BookE bolts the entire linear mapping */
483 	if (mmu_has_feature(MMU_FTR_TYPE_FSL_E))
484 		return linear_map_top;
485 	/* Other BookE, we assume the first GB is bolted */
486 	return 1ul << 30;
487 #else
488 	/* BookS, the first segment is bolted */
489 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
490 		return 1UL << SID_SHIFT_1T;
491 	return 1UL << SID_SHIFT;
492 #endif
493 }
494 
495 void __init irqstack_early_init(void)
496 {
497 	u64 limit = safe_stack_limit();
498 	unsigned int i;
499 
500 	/*
501 	 * Interrupt stacks must be in the first segment since we
502 	 * cannot afford to take SLB misses on them.
503 	 */
504 	for_each_possible_cpu(i) {
505 		softirq_ctx[i] = (struct thread_info *)
506 			__va(memblock_alloc_base(THREAD_SIZE,
507 					    THREAD_SIZE, limit));
508 		hardirq_ctx[i] = (struct thread_info *)
509 			__va(memblock_alloc_base(THREAD_SIZE,
510 					    THREAD_SIZE, limit));
511 	}
512 }
513 
514 #ifdef CONFIG_PPC_BOOK3E
515 void __init exc_lvl_early_init(void)
516 {
517 	unsigned int i;
518 	unsigned long sp;
519 
520 	for_each_possible_cpu(i) {
521 		sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
522 		critirq_ctx[i] = (struct thread_info *)__va(sp);
523 		paca[i].crit_kstack = __va(sp + THREAD_SIZE);
524 
525 		sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
526 		dbgirq_ctx[i] = (struct thread_info *)__va(sp);
527 		paca[i].dbg_kstack = __va(sp + THREAD_SIZE);
528 
529 		sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
530 		mcheckirq_ctx[i] = (struct thread_info *)__va(sp);
531 		paca[i].mc_kstack = __va(sp + THREAD_SIZE);
532 	}
533 
534 	if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC))
535 		patch_exception(0x040, exc_debug_debug_book3e);
536 }
537 #endif
538 
539 /*
540  * Stack space used when we detect a bad kernel stack pointer, and
541  * early in SMP boots before relocation is enabled. Exclusive emergency
542  * stack for machine checks.
543  */
544 void __init emergency_stack_init(void)
545 {
546 	u64 limit;
547 	unsigned int i;
548 
549 	/*
550 	 * Emergency stacks must be under 256MB, we cannot afford to take
551 	 * SLB misses on them. The ABI also requires them to be 128-byte
552 	 * aligned.
553 	 *
554 	 * Since we use these as temporary stacks during secondary CPU
555 	 * bringup, we need to get at them in real mode. This means they
556 	 * must also be within the RMO region.
557 	 */
558 	limit = min(safe_stack_limit(), ppc64_rma_size);
559 
560 	for_each_possible_cpu(i) {
561 		struct thread_info *ti;
562 		ti = __va(memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit));
563 		klp_init_thread_info(ti);
564 		paca[i].emergency_sp = (void *)ti + THREAD_SIZE;
565 
566 #ifdef CONFIG_PPC_BOOK3S_64
567 		/* emergency stack for machine check exception handling. */
568 		ti = __va(memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit));
569 		klp_init_thread_info(ti);
570 		paca[i].mc_emergency_sp = (void *)ti + THREAD_SIZE;
571 #endif
572 	}
573 }
574 
575 #ifdef CONFIG_SMP
576 #define PCPU_DYN_SIZE		()
577 
578 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
579 {
580 	return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
581 				    __pa(MAX_DMA_ADDRESS));
582 }
583 
584 static void __init pcpu_fc_free(void *ptr, size_t size)
585 {
586 	free_bootmem(__pa(ptr), size);
587 }
588 
589 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
590 {
591 	if (cpu_to_node(from) == cpu_to_node(to))
592 		return LOCAL_DISTANCE;
593 	else
594 		return REMOTE_DISTANCE;
595 }
596 
597 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
598 EXPORT_SYMBOL(__per_cpu_offset);
599 
600 void __init setup_per_cpu_areas(void)
601 {
602 	const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
603 	size_t atom_size;
604 	unsigned long delta;
605 	unsigned int cpu;
606 	int rc;
607 
608 	/*
609 	 * Linear mapping is one of 4K, 1M and 16M.  For 4K, no need
610 	 * to group units.  For larger mappings, use 1M atom which
611 	 * should be large enough to contain a number of units.
612 	 */
613 	if (mmu_linear_psize == MMU_PAGE_4K)
614 		atom_size = PAGE_SIZE;
615 	else
616 		atom_size = 1 << 20;
617 
618 	rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
619 				    pcpu_fc_alloc, pcpu_fc_free);
620 	if (rc < 0)
621 		panic("cannot initialize percpu area (err=%d)", rc);
622 
623 	delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
624 	for_each_possible_cpu(cpu) {
625                 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
626 		paca[cpu].data_offset = __per_cpu_offset[cpu];
627 	}
628 }
629 #endif
630 
631 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
632 unsigned long memory_block_size_bytes(void)
633 {
634 	if (ppc_md.memory_block_size)
635 		return ppc_md.memory_block_size();
636 
637 	return MIN_MEMORY_BLOCK_SIZE;
638 }
639 #endif
640 
641 #if defined(CONFIG_PPC_INDIRECT_PIO) || defined(CONFIG_PPC_INDIRECT_MMIO)
642 struct ppc_pci_io ppc_pci_io;
643 EXPORT_SYMBOL(ppc_pci_io);
644 #endif
645 
646 #ifdef CONFIG_HARDLOCKUP_DETECTOR
647 u64 hw_nmi_get_sample_period(int watchdog_thresh)
648 {
649 	return ppc_proc_freq * watchdog_thresh;
650 }
651 
652 /*
653  * The hardlockup detector breaks PMU event based branches and is likely
654  * to get false positives in KVM guests, so disable it by default.
655  */
656 static int __init disable_hardlockup_detector(void)
657 {
658 	hardlockup_detector_disable();
659 
660 	return 0;
661 }
662 early_initcall(disable_hardlockup_detector);
663 #endif
664