1 /*
2  * PPC64 code to handle Linux booting another kernel.
3  *
4  * Copyright (C) 2004-2005, IBM Corp.
5  *
6  * Created by: Milton D Miller II
7  *
8  * This source code is licensed under the GNU General Public License,
9  * Version 2.  See the file COPYING for more details.
10  */
11 
12 
13 #include <linux/kexec.h>
14 #include <linux/smp.h>
15 #include <linux/thread_info.h>
16 #include <linux/init_task.h>
17 #include <linux/errno.h>
18 #include <linux/kernel.h>
19 #include <linux/cpu.h>
20 
21 #include <asm/page.h>
22 #include <asm/current.h>
23 #include <asm/machdep.h>
24 #include <asm/cacheflush.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 
default_machine_kexec_prepare(struct kimage * image)32 int default_machine_kexec_prepare(struct kimage *image)
33 {
34 	int i;
35 	unsigned long begin, end;	/* limits of segment */
36 	unsigned long low, high;	/* limits of blocked memory range */
37 	struct device_node *node;
38 	const unsigned long *basep;
39 	const unsigned int *sizep;
40 
41 	if (!ppc_md.hpte_clear_all)
42 		return -ENOENT;
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 	/*
54 	 * For non-LPAR, we absolutely can not overwrite the mmu hash
55 	 * table, since we are still using the bolted entries in it to
56 	 * do the copy.  Check that here.
57 	 *
58 	 * It is safe if the end is below the start of the blocked
59 	 * region (end <= low), or if the beginning is after the
60 	 * end of the blocked region (begin >= high).  Use the
61 	 * boolean identity !(a || b)  === (!a && !b).
62 	 */
63 	if (htab_address) {
64 		low = __pa(htab_address);
65 		high = low + htab_size_bytes;
66 
67 		for (i = 0; i < image->nr_segments; i++) {
68 			begin = image->segment[i].mem;
69 			end = begin + image->segment[i].memsz;
70 
71 			if ((begin < high) && (end > low))
72 				return -ETXTBSY;
73 		}
74 	}
75 
76 	/* We also should not overwrite the tce tables */
77 	for_each_node_by_type(node, "pci") {
78 		basep = of_get_property(node, "linux,tce-base", NULL);
79 		sizep = of_get_property(node, "linux,tce-size", NULL);
80 		if (basep == NULL || sizep == NULL)
81 			continue;
82 
83 		low = *basep;
84 		high = low + (*sizep);
85 
86 		for (i = 0; i < image->nr_segments; i++) {
87 			begin = image->segment[i].mem;
88 			end = begin + image->segment[i].memsz;
89 
90 			if ((begin < high) && (end > low))
91 				return -ETXTBSY;
92 		}
93 	}
94 
95 	return 0;
96 }
97 
98 #define IND_FLAGS (IND_DESTINATION | IND_INDIRECTION | IND_DONE | IND_SOURCE)
99 
copy_segments(unsigned long ind)100 static void copy_segments(unsigned long ind)
101 {
102 	unsigned long entry;
103 	unsigned long *ptr;
104 	void *dest;
105 	void *addr;
106 
107 	/*
108 	 * We rely on kexec_load to create a lists that properly
109 	 * initializes these pointers before they are used.
110 	 * We will still crash if the list is wrong, but at least
111 	 * the compiler will be quiet.
112 	 */
113 	ptr = NULL;
114 	dest = NULL;
115 
116 	for (entry = ind; !(entry & IND_DONE); entry = *ptr++) {
117 		addr = __va(entry & PAGE_MASK);
118 
119 		switch (entry & IND_FLAGS) {
120 		case IND_DESTINATION:
121 			dest = addr;
122 			break;
123 		case IND_INDIRECTION:
124 			ptr = addr;
125 			break;
126 		case IND_SOURCE:
127 			copy_page(dest, addr);
128 			dest += PAGE_SIZE;
129 		}
130 	}
131 }
132 
kexec_copy_flush(struct kimage * image)133 void kexec_copy_flush(struct kimage *image)
134 {
135 	long i, nr_segments = image->nr_segments;
136 	struct  kexec_segment ranges[KEXEC_SEGMENT_MAX];
137 
138 	/* save the ranges on the stack to efficiently flush the icache */
139 	memcpy(ranges, image->segment, sizeof(ranges));
140 
141 	/*
142 	 * After this call we may not use anything allocated in dynamic
143 	 * memory, including *image.
144 	 *
145 	 * Only globals and the stack are allowed.
146 	 */
147 	copy_segments(image->head);
148 
149 	/*
150 	 * we need to clear the icache for all dest pages sometime,
151 	 * including ones that were in place on the original copy
152 	 */
153 	for (i = 0; i < nr_segments; i++)
154 		flush_icache_range((unsigned long)__va(ranges[i].mem),
155 			(unsigned long)__va(ranges[i].mem + ranges[i].memsz));
156 }
157 
158 #ifdef CONFIG_SMP
159 
160 static int kexec_all_irq_disabled = 0;
161 
kexec_smp_down(void * arg)162 static void kexec_smp_down(void *arg)
163 {
164 	local_irq_disable();
165 	hard_irq_disable();
166 
167 	mb(); /* make sure our irqs are disabled before we say they are */
168 	get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
169 	while(kexec_all_irq_disabled == 0)
170 		cpu_relax();
171 	mb(); /* make sure all irqs are disabled before this */
172 	hw_breakpoint_disable();
173 	/*
174 	 * Now every CPU has IRQs off, we can clear out any pending
175 	 * IPIs and be sure that no more will come in after this.
176 	 */
177 	if (ppc_md.kexec_cpu_down)
178 		ppc_md.kexec_cpu_down(0, 1);
179 
180 	kexec_smp_wait();
181 	/* NOTREACHED */
182 }
183 
kexec_prepare_cpus_wait(int wait_state)184 static void kexec_prepare_cpus_wait(int wait_state)
185 {
186 	int my_cpu, i, notified=-1;
187 
188 	hw_breakpoint_disable();
189 	my_cpu = get_cpu();
190 	/* Make sure each CPU has at least made it to the state we need.
191 	 *
192 	 * FIXME: There is a (slim) chance of a problem if not all of the CPUs
193 	 * are correctly onlined.  If somehow we start a CPU on boot with RTAS
194 	 * start-cpu, but somehow that CPU doesn't write callin_cpu_map[] in
195 	 * time, the boot CPU will timeout.  If it does eventually execute
196 	 * stuff, the secondary will start up (paca[].cpu_start was written) and
197 	 * get into a peculiar state.  If the platform supports
198 	 * smp_ops->take_timebase(), the secondary CPU will probably be spinning
199 	 * in there.  If not (i.e. pseries), the secondary will continue on and
200 	 * try to online itself/idle/etc. If it survives that, we need to find
201 	 * these possible-but-not-online-but-should-be CPUs and chaperone them
202 	 * into kexec_smp_wait().
203 	 */
204 	for_each_online_cpu(i) {
205 		if (i == my_cpu)
206 			continue;
207 
208 		while (paca[i].kexec_state < wait_state) {
209 			barrier();
210 			if (i != notified) {
211 				printk(KERN_INFO "kexec: waiting for cpu %d "
212 				       "(physical %d) to enter %i state\n",
213 				       i, paca[i].hw_cpu_id, wait_state);
214 				notified = i;
215 			}
216 		}
217 	}
218 	mb();
219 }
220 
221 /*
222  * We need to make sure each present CPU is online.  The next kernel will scan
223  * the device tree and assume primary threads are online and query secondary
224  * threads via RTAS to online them if required.  If we don't online primary
225  * threads, they will be stuck.  However, we also online secondary threads as we
226  * may be using 'cede offline'.  In this case RTAS doesn't see the secondary
227  * threads as offline -- and again, these CPUs will be stuck.
228  *
229  * So, we online all CPUs that should be running, including secondary threads.
230  */
wake_offline_cpus(void)231 static void wake_offline_cpus(void)
232 {
233 	int cpu = 0;
234 
235 	for_each_present_cpu(cpu) {
236 		if (!cpu_online(cpu)) {
237 			printk(KERN_INFO "kexec: Waking offline cpu %d.\n",
238 			       cpu);
239 			cpu_up(cpu);
240 		}
241 	}
242 }
243 
kexec_prepare_cpus(void)244 static void kexec_prepare_cpus(void)
245 {
246 	wake_offline_cpus();
247 	smp_call_function(kexec_smp_down, NULL, /* wait */0);
248 	local_irq_disable();
249 	hard_irq_disable();
250 
251 	mb(); /* make sure IRQs are disabled before we say they are */
252 	get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
253 
254 	kexec_prepare_cpus_wait(KEXEC_STATE_IRQS_OFF);
255 	/* we are sure every CPU has IRQs off at this point */
256 	kexec_all_irq_disabled = 1;
257 
258 	/* after we tell the others to go down */
259 	if (ppc_md.kexec_cpu_down)
260 		ppc_md.kexec_cpu_down(0, 0);
261 
262 	/*
263 	 * Before removing MMU mappings make sure all CPUs have entered real
264 	 * mode:
265 	 */
266 	kexec_prepare_cpus_wait(KEXEC_STATE_REAL_MODE);
267 
268 	put_cpu();
269 }
270 
271 #else /* ! SMP */
272 
kexec_prepare_cpus(void)273 static void kexec_prepare_cpus(void)
274 {
275 	/*
276 	 * move the secondarys to us so that we can copy
277 	 * the new kernel 0-0x100 safely
278 	 *
279 	 * do this if kexec in setup.c ?
280 	 *
281 	 * We need to release the cpus if we are ever going from an
282 	 * UP to an SMP kernel.
283 	 */
284 	smp_release_cpus();
285 	if (ppc_md.kexec_cpu_down)
286 		ppc_md.kexec_cpu_down(0, 0);
287 	local_irq_disable();
288 	hard_irq_disable();
289 }
290 
291 #endif /* SMP */
292 
293 /*
294  * kexec thread structure and stack.
295  *
296  * We need to make sure that this is 16384-byte aligned due to the
297  * way process stacks are handled.  It also must be statically allocated
298  * or allocated as part of the kimage, because everything else may be
299  * overwritten when we copy the kexec image.  We piggyback on the
300  * "init_task" linker section here to statically allocate a stack.
301  *
302  * We could use a smaller stack if we don't care about anything using
303  * current, but that audit has not been performed.
304  */
305 static union thread_union kexec_stack __init_task_data =
306 	{ };
307 
308 /*
309  * For similar reasons to the stack above, the kexecing CPU needs to be on a
310  * static PACA; we switch to kexec_paca.
311  */
312 struct paca_struct kexec_paca;
313 
314 /* Our assembly helper, in kexec_stub.S */
315 extern void kexec_sequence(void *newstack, unsigned long start,
316 			   void *image, void *control,
317 			   void (*clear_all)(void)) __noreturn;
318 
319 /* too late to fail here */
default_machine_kexec(struct kimage * image)320 void default_machine_kexec(struct kimage *image)
321 {
322 	/* prepare control code if any */
323 
324 	/*
325         * If the kexec boot is the normal one, need to shutdown other cpus
326         * into our wait loop and quiesce interrupts.
327         * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
328         * stopping other CPUs and collecting their pt_regs is done before
329         * using debugger IPI.
330         */
331 
332 	if (crashing_cpu == -1)
333 		kexec_prepare_cpus();
334 
335 	pr_debug("kexec: Starting switchover sequence.\n");
336 
337 	/* switch to a staticly allocated stack.  Based on irq stack code.
338 	 * XXX: the task struct will likely be invalid once we do the copy!
339 	 */
340 	kexec_stack.thread_info.task = current_thread_info()->task;
341 	kexec_stack.thread_info.flags = 0;
342 
343 	/* We need a static PACA, too; copy this CPU's PACA over and switch to
344 	 * it.  Also poison per_cpu_offset to catch anyone using non-static
345 	 * data.
346 	 */
347 	memcpy(&kexec_paca, get_paca(), sizeof(struct paca_struct));
348 	kexec_paca.data_offset = 0xedeaddeadeeeeeeeUL;
349 	paca = (struct paca_struct *)RELOC_HIDE(&kexec_paca, 0) -
350 		kexec_paca.paca_index;
351 	setup_paca(&kexec_paca);
352 
353 	/* XXX: If anyone does 'dynamic lppacas' this will also need to be
354 	 * switched to a static version!
355 	 */
356 
357 	/* Some things are best done in assembly.  Finding globals with
358 	 * a toc is easier in C, so pass in what we can.
359 	 */
360 	kexec_sequence(&kexec_stack, image->start, image,
361 			page_address(image->control_code_page),
362 			ppc_md.hpte_clear_all);
363 	/* NOTREACHED */
364 }
365 
366 /* Values we need to export to the second kernel via the device tree. */
367 static unsigned long htab_base;
368 
369 static struct property htab_base_prop = {
370 	.name = "linux,htab-base",
371 	.length = sizeof(unsigned long),
372 	.value = &htab_base,
373 };
374 
375 static struct property htab_size_prop = {
376 	.name = "linux,htab-size",
377 	.length = sizeof(unsigned long),
378 	.value = &htab_size_bytes,
379 };
380 
export_htab_values(void)381 static int __init export_htab_values(void)
382 {
383 	struct device_node *node;
384 	struct property *prop;
385 
386 	/* On machines with no htab htab_address is NULL */
387 	if (!htab_address)
388 		return -ENODEV;
389 
390 	node = of_find_node_by_path("/chosen");
391 	if (!node)
392 		return -ENODEV;
393 
394 	/* remove any stale propertys so ours can be found */
395 	prop = of_find_property(node, htab_base_prop.name, NULL);
396 	if (prop)
397 		prom_remove_property(node, prop);
398 	prop = of_find_property(node, htab_size_prop.name, NULL);
399 	if (prop)
400 		prom_remove_property(node, prop);
401 
402 	htab_base = __pa(htab_address);
403 	prom_add_property(node, &htab_base_prop);
404 	prom_add_property(node, &htab_size_prop);
405 
406 	of_node_put(node);
407 	return 0;
408 }
409 late_initcall(export_htab_values);
410