xref: /openbmc/linux/arch/powerpc/kexec/core_64.c (revision 35267cea)
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 
20 #include <asm/page.h>
21 #include <asm/current.h>
22 #include <asm/machdep.h>
23 #include <asm/cacheflush.h>
24 #include <asm/firmware.h>
25 #include <asm/paca.h>
26 #include <asm/mmu.h>
27 #include <asm/sections.h>	/* _end */
28 #include <asm/prom.h>
29 #include <asm/smp.h>
30 #include <asm/hw_breakpoint.h>
31 #include <asm/asm-prototypes.h>
32 #include <asm/svm.h>
33 #include <asm/ultravisor.h>
34 
35 int default_machine_kexec_prepare(struct kimage *image)
36 {
37 	int i;
38 	unsigned long begin, end;	/* limits of segment */
39 	unsigned long low, high;	/* limits of blocked memory range */
40 	struct device_node *node;
41 	const unsigned long *basep;
42 	const unsigned int *sizep;
43 
44 	/*
45 	 * Since we use the kernel fault handlers and paging code to
46 	 * handle the virtual mode, we must make sure no destination
47 	 * overlaps kernel static data or bss.
48 	 */
49 	for (i = 0; i < image->nr_segments; i++)
50 		if (image->segment[i].mem < __pa(_end))
51 			return -ETXTBSY;
52 
53 	/* We also should not overwrite the tce tables */
54 	for_each_node_by_type(node, "pci") {
55 		basep = of_get_property(node, "linux,tce-base", NULL);
56 		sizep = of_get_property(node, "linux,tce-size", NULL);
57 		if (basep == NULL || sizep == NULL)
58 			continue;
59 
60 		low = *basep;
61 		high = low + (*sizep);
62 
63 		for (i = 0; i < image->nr_segments; i++) {
64 			begin = image->segment[i].mem;
65 			end = begin + image->segment[i].memsz;
66 
67 			if ((begin < high) && (end > low))
68 				return -ETXTBSY;
69 		}
70 	}
71 
72 	return 0;
73 }
74 
75 static void copy_segments(unsigned long ind)
76 {
77 	unsigned long entry;
78 	unsigned long *ptr;
79 	void *dest;
80 	void *addr;
81 
82 	/*
83 	 * We rely on kexec_load to create a lists that properly
84 	 * initializes these pointers before they are used.
85 	 * We will still crash if the list is wrong, but at least
86 	 * the compiler will be quiet.
87 	 */
88 	ptr = NULL;
89 	dest = NULL;
90 
91 	for (entry = ind; !(entry & IND_DONE); entry = *ptr++) {
92 		addr = __va(entry & PAGE_MASK);
93 
94 		switch (entry & IND_FLAGS) {
95 		case IND_DESTINATION:
96 			dest = addr;
97 			break;
98 		case IND_INDIRECTION:
99 			ptr = addr;
100 			break;
101 		case IND_SOURCE:
102 			copy_page(dest, addr);
103 			dest += PAGE_SIZE;
104 		}
105 	}
106 }
107 
108 void kexec_copy_flush(struct kimage *image)
109 {
110 	long i, nr_segments = image->nr_segments;
111 	struct  kexec_segment ranges[KEXEC_SEGMENT_MAX];
112 
113 	/* save the ranges on the stack to efficiently flush the icache */
114 	memcpy(ranges, image->segment, sizeof(ranges));
115 
116 	/*
117 	 * After this call we may not use anything allocated in dynamic
118 	 * memory, including *image.
119 	 *
120 	 * Only globals and the stack are allowed.
121 	 */
122 	copy_segments(image->head);
123 
124 	/*
125 	 * we need to clear the icache for all dest pages sometime,
126 	 * including ones that were in place on the original copy
127 	 */
128 	for (i = 0; i < nr_segments; i++)
129 		flush_icache_range((unsigned long)__va(ranges[i].mem),
130 			(unsigned long)__va(ranges[i].mem + ranges[i].memsz));
131 }
132 
133 #ifdef CONFIG_SMP
134 
135 static int kexec_all_irq_disabled = 0;
136 
137 static void kexec_smp_down(void *arg)
138 {
139 	local_irq_disable();
140 	hard_irq_disable();
141 
142 	mb(); /* make sure our irqs are disabled before we say they are */
143 	get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
144 	while(kexec_all_irq_disabled == 0)
145 		cpu_relax();
146 	mb(); /* make sure all irqs are disabled before this */
147 	hw_breakpoint_disable();
148 	/*
149 	 * Now every CPU has IRQs off, we can clear out any pending
150 	 * IPIs and be sure that no more will come in after this.
151 	 */
152 	if (ppc_md.kexec_cpu_down)
153 		ppc_md.kexec_cpu_down(0, 1);
154 
155 	reset_sprs();
156 
157 	kexec_smp_wait();
158 	/* NOTREACHED */
159 }
160 
161 static void kexec_prepare_cpus_wait(int wait_state)
162 {
163 	int my_cpu, i, notified=-1;
164 
165 	hw_breakpoint_disable();
166 	my_cpu = get_cpu();
167 	/* Make sure each CPU has at least made it to the state we need.
168 	 *
169 	 * FIXME: There is a (slim) chance of a problem if not all of the CPUs
170 	 * are correctly onlined.  If somehow we start a CPU on boot with RTAS
171 	 * start-cpu, but somehow that CPU doesn't write callin_cpu_map[] in
172 	 * time, the boot CPU will timeout.  If it does eventually execute
173 	 * stuff, the secondary will start up (paca_ptrs[]->cpu_start was
174 	 * written) and get into a peculiar state.
175 	 * If the platform supports smp_ops->take_timebase(), the secondary CPU
176 	 * will probably be spinning in there.  If not (i.e. pseries), the
177 	 * secondary will continue on and try to online itself/idle/etc. If it
178 	 * survives that, we need to find these
179 	 * possible-but-not-online-but-should-be CPUs and chaperone them into
180 	 * kexec_smp_wait().
181 	 */
182 	for_each_online_cpu(i) {
183 		if (i == my_cpu)
184 			continue;
185 
186 		while (paca_ptrs[i]->kexec_state < wait_state) {
187 			barrier();
188 			if (i != notified) {
189 				printk(KERN_INFO "kexec: waiting for cpu %d "
190 				       "(physical %d) to enter %i state\n",
191 				       i, paca_ptrs[i]->hw_cpu_id, wait_state);
192 				notified = i;
193 			}
194 		}
195 	}
196 	mb();
197 }
198 
199 /*
200  * We need to make sure each present CPU is online.  The next kernel will scan
201  * the device tree and assume primary threads are online and query secondary
202  * threads via RTAS to online them if required.  If we don't online primary
203  * threads, they will be stuck.  However, we also online secondary threads as we
204  * may be using 'cede offline'.  In this case RTAS doesn't see the secondary
205  * threads as offline -- and again, these CPUs will be stuck.
206  *
207  * So, we online all CPUs that should be running, including secondary threads.
208  */
209 static void wake_offline_cpus(void)
210 {
211 	int cpu = 0;
212 
213 	for_each_present_cpu(cpu) {
214 		if (!cpu_online(cpu)) {
215 			printk(KERN_INFO "kexec: Waking offline cpu %d.\n",
216 			       cpu);
217 			WARN_ON(add_cpu(cpu));
218 		}
219 	}
220 }
221 
222 static void kexec_prepare_cpus(void)
223 {
224 	wake_offline_cpus();
225 	smp_call_function(kexec_smp_down, NULL, /* wait */0);
226 	local_irq_disable();
227 	hard_irq_disable();
228 
229 	mb(); /* make sure IRQs are disabled before we say they are */
230 	get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
231 
232 	kexec_prepare_cpus_wait(KEXEC_STATE_IRQS_OFF);
233 	/* we are sure every CPU has IRQs off at this point */
234 	kexec_all_irq_disabled = 1;
235 
236 	/*
237 	 * Before removing MMU mappings make sure all CPUs have entered real
238 	 * mode:
239 	 */
240 	kexec_prepare_cpus_wait(KEXEC_STATE_REAL_MODE);
241 
242 	/* after we tell the others to go down */
243 	if (ppc_md.kexec_cpu_down)
244 		ppc_md.kexec_cpu_down(0, 0);
245 
246 	put_cpu();
247 }
248 
249 #else /* ! SMP */
250 
251 static void kexec_prepare_cpus(void)
252 {
253 	/*
254 	 * move the secondarys to us so that we can copy
255 	 * the new kernel 0-0x100 safely
256 	 *
257 	 * do this if kexec in setup.c ?
258 	 *
259 	 * We need to release the cpus if we are ever going from an
260 	 * UP to an SMP kernel.
261 	 */
262 	smp_release_cpus();
263 	if (ppc_md.kexec_cpu_down)
264 		ppc_md.kexec_cpu_down(0, 0);
265 	local_irq_disable();
266 	hard_irq_disable();
267 }
268 
269 #endif /* SMP */
270 
271 /*
272  * kexec thread structure and stack.
273  *
274  * We need to make sure that this is 16384-byte aligned due to the
275  * way process stacks are handled.  It also must be statically allocated
276  * or allocated as part of the kimage, because everything else may be
277  * overwritten when we copy the kexec image.  We piggyback on the
278  * "init_task" linker section here to statically allocate a stack.
279  *
280  * We could use a smaller stack if we don't care about anything using
281  * current, but that audit has not been performed.
282  */
283 static union thread_union kexec_stack __init_task_data =
284 	{ };
285 
286 /*
287  * For similar reasons to the stack above, the kexecing CPU needs to be on a
288  * static PACA; we switch to kexec_paca.
289  */
290 struct paca_struct kexec_paca;
291 
292 /* Our assembly helper, in misc_64.S */
293 extern void kexec_sequence(void *newstack, unsigned long start,
294 			   void *image, void *control,
295 			   void (*clear_all)(void),
296 			   bool copy_with_mmu_off) __noreturn;
297 
298 /* too late to fail here */
299 void default_machine_kexec(struct kimage *image)
300 {
301 	bool copy_with_mmu_off;
302 
303 	/* prepare control code if any */
304 
305 	/*
306         * If the kexec boot is the normal one, need to shutdown other cpus
307         * into our wait loop and quiesce interrupts.
308         * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
309         * stopping other CPUs and collecting their pt_regs is done before
310         * using debugger IPI.
311         */
312 
313 	if (!kdump_in_progress())
314 		kexec_prepare_cpus();
315 
316 	printk("kexec: Starting switchover sequence.\n");
317 
318 	/* switch to a staticly allocated stack.  Based on irq stack code.
319 	 * We setup preempt_count to avoid using VMX in memcpy.
320 	 * XXX: the task struct will likely be invalid once we do the copy!
321 	 */
322 	current_thread_info()->flags = 0;
323 	current_thread_info()->preempt_count = HARDIRQ_OFFSET;
324 
325 	/* We need a static PACA, too; copy this CPU's PACA over and switch to
326 	 * it. Also poison per_cpu_offset and NULL lppaca to catch anyone using
327 	 * non-static data.
328 	 */
329 	memcpy(&kexec_paca, get_paca(), sizeof(struct paca_struct));
330 	kexec_paca.data_offset = 0xedeaddeadeeeeeeeUL;
331 #ifdef CONFIG_PPC_PSERIES
332 	kexec_paca.lppaca_ptr = NULL;
333 #endif
334 
335 	if (is_secure_guest() && !(image->preserve_context ||
336 				   image->type == KEXEC_TYPE_CRASH)) {
337 		uv_unshare_all_pages();
338 		printk("kexec: Unshared all shared pages.\n");
339 	}
340 
341 	paca_ptrs[kexec_paca.paca_index] = &kexec_paca;
342 
343 	setup_paca(&kexec_paca);
344 
345 	/*
346 	 * The lppaca should be unregistered at this point so the HV won't
347 	 * touch it. In the case of a crash, none of the lppacas are
348 	 * unregistered so there is not much we can do about it here.
349 	 */
350 
351 	/*
352 	 * On Book3S, the copy must happen with the MMU off if we are either
353 	 * using Radix page tables or we are not in an LPAR since we can
354 	 * overwrite the page tables while copying.
355 	 *
356 	 * In an LPAR, we keep the MMU on otherwise we can't access beyond
357 	 * the RMA. On BookE there is no real MMU off mode, so we have to
358 	 * keep it enabled as well (but then we have bolted TLB entries).
359 	 */
360 #ifdef CONFIG_PPC_BOOK3E
361 	copy_with_mmu_off = false;
362 #else
363 	copy_with_mmu_off = radix_enabled() ||
364 		!(firmware_has_feature(FW_FEATURE_LPAR) ||
365 		  firmware_has_feature(FW_FEATURE_PS3_LV1));
366 #endif
367 
368 	/* Some things are best done in assembly.  Finding globals with
369 	 * a toc is easier in C, so pass in what we can.
370 	 */
371 	kexec_sequence(&kexec_stack, image->start, image,
372 		       page_address(image->control_code_page),
373 		       mmu_cleanup_all, copy_with_mmu_off);
374 	/* NOTREACHED */
375 }
376 
377 #ifdef CONFIG_PPC_BOOK3S_64
378 /* Values we need to export to the second kernel via the device tree. */
379 static unsigned long htab_base;
380 static unsigned long htab_size;
381 
382 static struct property htab_base_prop = {
383 	.name = "linux,htab-base",
384 	.length = sizeof(unsigned long),
385 	.value = &htab_base,
386 };
387 
388 static struct property htab_size_prop = {
389 	.name = "linux,htab-size",
390 	.length = sizeof(unsigned long),
391 	.value = &htab_size,
392 };
393 
394 static int __init export_htab_values(void)
395 {
396 	struct device_node *node;
397 
398 	/* On machines with no htab htab_address is NULL */
399 	if (!htab_address)
400 		return -ENODEV;
401 
402 	node = of_find_node_by_path("/chosen");
403 	if (!node)
404 		return -ENODEV;
405 
406 	/* remove any stale propertys so ours can be found */
407 	of_remove_property(node, of_find_property(node, htab_base_prop.name, NULL));
408 	of_remove_property(node, of_find_property(node, htab_size_prop.name, NULL));
409 
410 	htab_base = cpu_to_be64(__pa(htab_address));
411 	of_add_property(node, &htab_base_prop);
412 	htab_size = cpu_to_be64(htab_size_bytes);
413 	of_add_property(node, &htab_size_prop);
414 
415 	of_node_put(node);
416 	return 0;
417 }
418 late_initcall(export_htab_values);
419 #endif /* CONFIG_PPC_BOOK3S_64 */
420