core_64.c

  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * PPC64 code to handle Linux booting another kernel.
  4 *
  5 * Copyright (C) 2004-2005, IBM Corp.
  6 *
  7 * Created by: Milton D Miller II
  8 */
  9
 10
 11#include <linux/kexec.h>
 12#include <linux/smp.h>
 13#include <linux/thread_info.h>
 14#include <linux/init_task.h>
 15#include <linux/errno.h>
 16#include <linux/kernel.h>
 17#include <linux/cpu.h>
 18#include <linux/hardirq.h>
 19#include <linux/of.h>
 20#include <linux/libfdt.h>
 21
 22#include <asm/page.h>
 23#include <asm/current.h>
 24#include <asm/machdep.h>
 25#include <asm/cacheflush.h>
 26#include <asm/firmware.h>
 27#include <asm/paca.h>
 28#include <asm/mmu.h>
 29#include <asm/sections.h>	/* _end */
 30#include <asm/setup.h>
 31#include <asm/smp.h>
 32#include <asm/hw_breakpoint.h>
 33#include <asm/svm.h>
 34#include <asm/ultravisor.h>
 35#include <asm/crashdump-ppc64.h>
 36
 37int machine_kexec_prepare(struct kimage *image)
 38{
 39	int i;
 40	unsigned long begin, end;	/* limits of segment */
 41	unsigned long low, high;	/* limits of blocked memory range */
 42	struct device_node *node;
 43	const unsigned long *basep;
 44	const unsigned int *sizep;
 45
 46	/*
 47	 * Since we use the kernel fault handlers and paging code to
 48	 * handle the virtual mode, we must make sure no destination
 49	 * overlaps kernel static data or bss.
 50	 */
 51	for (i = 0; i < image->nr_segments; i++)
 52		if (image->segment[i].mem < __pa(_end))
 53			return -ETXTBSY;
 54
 55	/* We also should not overwrite the tce tables */
 56	for_each_node_by_type(node, "pci") {
 57		basep = of_get_property(node, "linux,tce-base", NULL);
 58		sizep = of_get_property(node, "linux,tce-size", NULL);
 59		if (basep == NULL || sizep == NULL)
 60			continue;
 61
 62		low = *basep;
 63		high = low + (*sizep);
 64
 65		for (i = 0; i < image->nr_segments; i++) {
 66			begin = image->segment[i].mem;
 67			end = begin + image->segment[i].memsz;
 68
 69			if ((begin < high) && (end > low)) {
 70				of_node_put(node);
 71				return -ETXTBSY;
 72			}
 73		}
 74	}
 75
 76	return 0;
 77}
 78
 79/* Called during kexec sequence with MMU off */
 80static notrace void copy_segments(unsigned long ind)
 81{
 82	unsigned long entry;
 83	unsigned long *ptr;
 84	void *dest;
 85	void *addr;
 86
 87	/*
 88	 * We rely on kexec_load to create a lists that properly
 89	 * initializes these pointers before they are used.
 90	 * We will still crash if the list is wrong, but at least
 91	 * the compiler will be quiet.
 92	 */
 93	ptr = NULL;
 94	dest = NULL;
 95
 96	for (entry = ind; !(entry & IND_DONE); entry = *ptr++) {
 97		addr = __va(entry & PAGE_MASK);
 98
 99		switch (entry & IND_FLAGS) {
100		case IND_DESTINATION:
101			dest = addr;
102			break;
103		case IND_INDIRECTION:
104			ptr = addr;
105			break;
106		case IND_SOURCE:
107			copy_page(dest, addr);
108			dest += PAGE_SIZE;
109		}
110	}
111}
112
113/* Called during kexec sequence with MMU off */
114notrace void kexec_copy_flush(struct kimage *image)
115{
116	long i, nr_segments = image->nr_segments;
117	struct  kexec_segment ranges[KEXEC_SEGMENT_MAX];
118
119	/* save the ranges on the stack to efficiently flush the icache */
120	memcpy(ranges, image->segment, sizeof(ranges));
121
122	/*
123	 * After this call we may not use anything allocated in dynamic
124	 * memory, including *image.
125	 *
126	 * Only globals and the stack are allowed.
127	 */
128	copy_segments(image->head);
129
130	/*
131	 * we need to clear the icache for all dest pages sometime,
132	 * including ones that were in place on the original copy
133	 */
134	for (i = 0; i < nr_segments; i++)
135		flush_icache_range((unsigned long)__va(ranges[i].mem),
136			(unsigned long)__va(ranges[i].mem + ranges[i].memsz));
137}
138
139#ifdef CONFIG_SMP
140
141static int kexec_all_irq_disabled = 0;
142
143static void kexec_smp_down(void *arg)
144{
145	local_irq_disable();
146	hard_irq_disable();
147
148	mb(); /* make sure our irqs are disabled before we say they are */
149	get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
150	while(kexec_all_irq_disabled == 0)
151		cpu_relax();
152	mb(); /* make sure all irqs are disabled before this */
153	hw_breakpoint_disable();
154	/*
155	 * Now every CPU has IRQs off, we can clear out any pending
156	 * IPIs and be sure that no more will come in after this.
157	 */
158	if (ppc_md.kexec_cpu_down)
159		ppc_md.kexec_cpu_down(0, 1);
160
161	reset_sprs();
162
163	kexec_smp_wait();
164	/* NOTREACHED */
165}
166
167static void kexec_prepare_cpus_wait(int wait_state)
168{
169	int my_cpu, i, notified=-1;
170
171	hw_breakpoint_disable();
172	my_cpu = get_cpu();
173	/* Make sure each CPU has at least made it to the state we need.
174	 *
175	 * FIXME: There is a (slim) chance of a problem if not all of the CPUs
176	 * are correctly onlined.  If somehow we start a CPU on boot with RTAS
177	 * start-cpu, but somehow that CPU doesn't write callin_cpu_map[] in
178	 * time, the boot CPU will timeout.  If it does eventually execute
179	 * stuff, the secondary will start up (paca_ptrs[]->cpu_start was
180	 * written) and get into a peculiar state.
181	 * If the platform supports smp_ops->take_timebase(), the secondary CPU
182	 * will probably be spinning in there.  If not (i.e. pseries), the
183	 * secondary will continue on and try to online itself/idle/etc. If it
184	 * survives that, we need to find these
185	 * possible-but-not-online-but-should-be CPUs and chaperone them into
186	 * kexec_smp_wait().
187	 */
188	for_each_online_cpu(i) {
189		if (i == my_cpu)
190			continue;
191
192		while (paca_ptrs[i]->kexec_state < wait_state) {
193			barrier();
194			if (i != notified) {
195				printk(KERN_INFO "kexec: waiting for cpu %d "
196				       "(physical %d) to enter %i state\n",
197				       i, paca_ptrs[i]->hw_cpu_id, wait_state);
198				notified = i;
199			}
200		}
201	}
202	mb();
203}
204
205
206/*
207 * The add_cpu() call in wake_offline_cpus() can fail as cpu_bootable()
208 * returns false for CPUs that fail the cpu_smt_thread_allowed() check
209 * or non primary threads if SMT is disabled. Re-enable SMT and set the
210 * number of SMT threads to threads per core.
211 */
212static void kexec_smt_reenable(void)
213{
214#if defined(CONFIG_SMP) && defined(CONFIG_HOTPLUG_SMT)
215	lock_device_hotplug();
216	cpu_smt_num_threads = threads_per_core;
217	cpu_smt_control = CPU_SMT_ENABLED;
218	unlock_device_hotplug();
219#endif
220}
221
222/*
223 * We need to make sure each present CPU is online.  The next kernel will scan
224 * the device tree and assume primary threads are online and query secondary
225 * threads via RTAS to online them if required.  If we don't online primary
226 * threads, they will be stuck.  However, we also online secondary threads as we
227 * may be using 'cede offline'.  In this case RTAS doesn't see the secondary
228 * threads as offline -- and again, these CPUs will be stuck.
229 *
230 * So, we online all CPUs that should be running, including secondary threads.
231 */
232static void wake_offline_cpus(void)
233{
234	int cpu = 0;
235
236	kexec_smt_reenable();
237
238	for_each_present_cpu(cpu) {
239		if (!cpu_online(cpu)) {
240			printk(KERN_INFO "kexec: Waking offline cpu %d.\n",
241			       cpu);
242			WARN_ON(add_cpu(cpu));
243		}
244	}
245}
246
247static void kexec_prepare_cpus(void)
248{
249	wake_offline_cpus();
250	smp_call_function(kexec_smp_down, NULL, /* wait */0);
251	local_irq_disable();
252	hard_irq_disable();
253
254	mb(); /* make sure IRQs are disabled before we say they are */
255	get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
256
257	kexec_prepare_cpus_wait(KEXEC_STATE_IRQS_OFF);
258	/* we are sure every CPU has IRQs off at this point */
259	kexec_all_irq_disabled = 1;
260
261	/*
262	 * Before removing MMU mappings make sure all CPUs have entered real
263	 * mode:
264	 */
265	kexec_prepare_cpus_wait(KEXEC_STATE_REAL_MODE);
266
267	/* after we tell the others to go down */
268	if (ppc_md.kexec_cpu_down)
269		ppc_md.kexec_cpu_down(0, 0);
270
271	put_cpu();
272}
273
274#else /* ! SMP */
275
276static void kexec_prepare_cpus(void)
277{
278	/*
279	 * move the secondarys to us so that we can copy
280	 * the new kernel 0-0x100 safely
281	 *
282	 * do this if kexec in setup.c ?
283	 *
284	 * We need to release the cpus if we are ever going from an
285	 * UP to an SMP kernel.
286	 */
287	smp_release_cpus();
288	if (ppc_md.kexec_cpu_down)
289		ppc_md.kexec_cpu_down(0, 0);
290	local_irq_disable();
291	hard_irq_disable();
292}
293
294#endif /* SMP */
295
296/*
297 * kexec thread structure and stack.
298 *
299 * We need to make sure that this is 16384-byte aligned due to the
300 * way process stacks are handled.  It also must be statically allocated
301 * or allocated as part of the kimage, because everything else may be
302 * overwritten when we copy the kexec image.  We piggyback on the
303 * "init_task" linker section here to statically allocate a stack.
304 *
305 * We could use a smaller stack if we don't care about anything using
306 * current, but that audit has not been performed.
307 */
308static union thread_union kexec_stack = { };
309
310/*
311 * For similar reasons to the stack above, the kexecing CPU needs to be on a
312 * static PACA; we switch to kexec_paca.
313 */
314static struct paca_struct kexec_paca;
315
316/* Our assembly helper, in misc_64.S */
317extern void kexec_sequence(void *newstack, unsigned long start,
318			   void *image, void *control,
319			   void (*clear_all)(void),
320			   bool copy_with_mmu_off) __noreturn;
321
322/* too late to fail here */
323void default_machine_kexec(struct kimage *image)
324{
325	bool copy_with_mmu_off;
326
327	/* prepare control code if any */
328
329	/*
330        * If the kexec boot is the normal one, need to shutdown other cpus
331        * into our wait loop and quiesce interrupts.
332        * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
333        * stopping other CPUs and collecting their pt_regs is done before
334        * using debugger IPI.
335        */
336
337	if (!kdump_in_progress())
338		kexec_prepare_cpus();
339
340#ifdef CONFIG_PPC_PSERIES
341	/*
342	 * This must be done after other CPUs have shut down, otherwise they
343	 * could execute the 'scv' instruction, which is not supported with
344	 * reloc disabled (see configure_exceptions()).
345	 */
346	if (firmware_has_feature(FW_FEATURE_SET_MODE))
347		pseries_disable_reloc_on_exc();
348#endif
349
350	printk("kexec: Starting switchover sequence.\n");
351
352	/* switch to a staticly allocated stack.  Based on irq stack code.
353	 * We setup preempt_count to avoid using VMX in memcpy.
354	 * XXX: the task struct will likely be invalid once we do the copy!
355	 */
356	current_thread_info()->flags = 0;
357	current_thread_info()->preempt_count = HARDIRQ_OFFSET;
358
359	/* We need a static PACA, too; copy this CPU's PACA over and switch to
360	 * it. Also poison per_cpu_offset and NULL lppaca to catch anyone using
361	 * non-static data.
362	 */
363	memcpy(&kexec_paca, get_paca(), sizeof(struct paca_struct));
364	kexec_paca.data_offset = 0xedeaddeadeeeeeeeUL;
365#ifdef CONFIG_PPC_PSERIES
366	kexec_paca.lppaca_ptr = NULL;
367#endif
368
369	if (is_secure_guest() && !(image->preserve_context ||
370				   image->type == KEXEC_TYPE_CRASH)) {
371		uv_unshare_all_pages();
372		printk("kexec: Unshared all shared pages.\n");
373	}
374
375	paca_ptrs[kexec_paca.paca_index] = &kexec_paca;
376
377	setup_paca(&kexec_paca);
378
379	/*
380	 * The lppaca should be unregistered at this point so the HV won't
381	 * touch it. In the case of a crash, none of the lppacas are
382	 * unregistered so there is not much we can do about it here.
383	 */
384
385	/*
386	 * On Book3S, the copy must happen with the MMU off if we are either
387	 * using Radix page tables or we are not in an LPAR since we can
388	 * overwrite the page tables while copying.
389	 *
390	 * In an LPAR, we keep the MMU on otherwise we can't access beyond
391	 * the RMA. On BookE there is no real MMU off mode, so we have to
392	 * keep it enabled as well (but then we have bolted TLB entries).
393	 */
394#ifdef CONFIG_PPC_BOOK3E_64
395	copy_with_mmu_off = false;
396#else
397	copy_with_mmu_off = radix_enabled() ||
398		!(firmware_has_feature(FW_FEATURE_LPAR) ||
399		  firmware_has_feature(FW_FEATURE_PS3_LV1));
400#endif
401
402	/* Some things are best done in assembly.  Finding globals with
403	 * a toc is easier in C, so pass in what we can.
404	 */
405	kexec_sequence(&kexec_stack, image->start, image,
406		       page_address(image->control_code_page),
407		       mmu_cleanup_all, copy_with_mmu_off);
408	/* NOTREACHED */
409}
410
411#ifdef CONFIG_PPC_64S_HASH_MMU
412/* Values we need to export to the second kernel via the device tree. */
413static __be64 htab_base;
414static __be64 htab_size;
415
416static struct property htab_base_prop = {
417	.name = "linux,htab-base",
418	.length = sizeof(unsigned long),
419	.value = &htab_base,
420};
421
422static struct property htab_size_prop = {
423	.name = "linux,htab-size",
424	.length = sizeof(unsigned long),
425	.value = &htab_size,
426};
427
428static int __init export_htab_values(void)
429{
430	struct device_node *node;
431
432	/* On machines with no htab htab_address is NULL */
433	if (!htab_address)
434		return -ENODEV;
435
436	node = of_find_node_by_path("/chosen");
437	if (!node)
438		return -ENODEV;
439
440	/* remove any stale properties so ours can be found */
441	of_remove_property(node, of_find_property(node, htab_base_prop.name, NULL));
442	of_remove_property(node, of_find_property(node, htab_size_prop.name, NULL));
443
444	htab_base = cpu_to_be64(__pa(htab_address));
445	of_add_property(node, &htab_base_prop);
446	htab_size = cpu_to_be64(htab_size_bytes);
447	of_add_property(node, &htab_size_prop);
448
449	of_node_put(node);
450	return 0;
451}
452late_initcall(export_htab_values);
453#endif /* CONFIG_PPC_64S_HASH_MMU */
454
455#if defined(CONFIG_KEXEC_FILE) || defined(CONFIG_CRASH_DUMP)
456/**
457 * add_node_props - Reads node properties from device node structure and add
458 *                  them to fdt.
459 * @fdt:            Flattened device tree of the kernel
460 * @node_offset:    offset of the node to add a property at
461 * @dn:             device node pointer
462 *
463 * Returns 0 on success, negative errno on error.
464 */
465static int add_node_props(void *fdt, int node_offset, const struct device_node *dn)
466{
467	int ret = 0;
468	struct property *pp;
469
470	if (!dn)
471		return -EINVAL;
472
473	for_each_property_of_node(dn, pp) {
474		ret = fdt_setprop(fdt, node_offset, pp->name, pp->value, pp->length);
475		if (ret < 0) {
476			pr_err("Unable to add %s property: %s\n", pp->name, fdt_strerror(ret));
477			return ret;
478		}
479	}
480	return ret;
481}
482
483/**
484 * update_cpus_node - Update cpus node of flattened device tree using of_root
485 *                    device node.
486 * @fdt:              Flattened device tree of the kernel.
487 *
488 * Returns 0 on success, negative errno on error.
489 *
490 * Note: expecting no subnodes under /cpus/<node> with device_type == "cpu".
491 * If this changes, update this function to include them.
492 */
493int update_cpus_node(void *fdt)
494{
495	int prev_node_offset;
496	const char *device_type;
497	const struct fdt_property *prop;
498	struct device_node *cpus_node, *dn;
499	int cpus_offset, cpus_subnode_offset, ret = 0;
500
501	cpus_offset = fdt_path_offset(fdt, "/cpus");
502	if (cpus_offset < 0 && cpus_offset != -FDT_ERR_NOTFOUND) {
503		pr_err("Malformed device tree: error reading /cpus node: %s\n",
504		       fdt_strerror(cpus_offset));
505		return cpus_offset;
506	}
507
508	prev_node_offset = cpus_offset;
509	/* Delete sub-nodes of /cpus node with device_type == "cpu" */
510	for (cpus_subnode_offset = fdt_first_subnode(fdt, cpus_offset); cpus_subnode_offset >= 0;) {
511		/* Ignore nodes that do not have a device_type property or device_type != "cpu" */
512		prop = fdt_get_property(fdt, cpus_subnode_offset, "device_type", NULL);
513		if (!prop || strcmp(prop->data, "cpu")) {
514			prev_node_offset = cpus_subnode_offset;
515			goto next_node;
516		}
517
518		ret = fdt_del_node(fdt, cpus_subnode_offset);
519		if (ret < 0) {
520			pr_err("Failed to delete a cpus sub-node: %s\n", fdt_strerror(ret));
521			return ret;
522		}
523next_node:
524		if (prev_node_offset == cpus_offset)
525			cpus_subnode_offset = fdt_first_subnode(fdt, cpus_offset);
526		else
527			cpus_subnode_offset = fdt_next_subnode(fdt, prev_node_offset);
528	}
529
530	cpus_node = of_find_node_by_path("/cpus");
531	/* Fail here to avoid kexec/kdump kernel boot hung */
532	if (!cpus_node) {
533		pr_err("No /cpus node found\n");
534		return -EINVAL;
535	}
536
537	/* Add all /cpus sub-nodes of device_type == "cpu" to FDT */
538	for_each_child_of_node(cpus_node, dn) {
539		/* Ignore device nodes that do not have a device_type property
540		 * or device_type != "cpu".
541		 */
542		device_type = of_get_property(dn, "device_type", NULL);
543		if (!device_type || strcmp(device_type, "cpu"))
544			continue;
545
546		cpus_subnode_offset = fdt_add_subnode(fdt, cpus_offset, dn->full_name);
547		if (cpus_subnode_offset < 0) {
548			pr_err("Unable to add %s subnode: %s\n", dn->full_name,
549			       fdt_strerror(cpus_subnode_offset));
550			ret = cpus_subnode_offset;
551			goto out;
552		}
553
554		ret = add_node_props(fdt, cpus_subnode_offset, dn);
555		if (ret < 0)
556			goto out;
557	}
558out:
559	of_node_put(cpus_node);
560	of_node_put(dn);
561	return ret;
562}
563#endif /* CONFIG_KEXEC_FILE || CONFIG_CRASH_DUMP */