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#include <linux/hardirq.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/paca.h>
27#include <asm/mmu.h>
28#include <asm/sections.h>	/* _end */
29#include <asm/prom.h>
30#include <asm/smp.h>
31#include <asm/hw_breakpoint.h>
32
33#ifdef CONFIG_PPC_BOOK3E
34int default_machine_kexec_prepare(struct kimage *image)
35{
36	int i;
37	/*
38	 * Since we use the kernel fault handlers and paging code to
39	 * handle the virtual mode, we must make sure no destination
40	 * overlaps kernel static data or bss.
41	 */
42	for (i = 0; i < image->nr_segments; i++)
43		if (image->segment[i].mem < __pa(_end))
44			return -ETXTBSY;
45	return 0;
46}
47#else
48int default_machine_kexec_prepare(struct kimage *image)
49{
50	int i;
51	unsigned long begin, end;	/* limits of segment */
52	unsigned long low, high;	/* limits of blocked memory range */
53	struct device_node *node;
54	const unsigned long *basep;
55	const unsigned int *sizep;
56
57	if (!ppc_md.hpte_clear_all)
58		return -ENOENT;
59
60	/*
61	 * Since we use the kernel fault handlers and paging code to
62	 * handle the virtual mode, we must make sure no destination
63	 * overlaps kernel static data or bss.
64	 */
65	for (i = 0; i < image->nr_segments; i++)
66		if (image->segment[i].mem < __pa(_end))
67			return -ETXTBSY;
68
69	/*
70	 * For non-LPAR, we absolutely can not overwrite the mmu hash
71	 * table, since we are still using the bolted entries in it to
72	 * do the copy.  Check that here.
73	 *
74	 * It is safe if the end is below the start of the blocked
75	 * region (end <= low), or if the beginning is after the
76	 * end of the blocked region (begin >= high).  Use the
77	 * boolean identity !(a || b)  === (!a && !b).
78	 */
79	if (htab_address) {
80		low = __pa(htab_address);
81		high = low + htab_size_bytes;
82
83		for (i = 0; i < image->nr_segments; i++) {
84			begin = image->segment[i].mem;
85			end = begin + image->segment[i].memsz;
86
87			if ((begin < high) && (end > low))
88				return -ETXTBSY;
89		}
90	}
91
92	/* We also should not overwrite the tce tables */
93	for_each_node_by_type(node, "pci") {
94		basep = of_get_property(node, "linux,tce-base", NULL);
95		sizep = of_get_property(node, "linux,tce-size", NULL);
96		if (basep == NULL || sizep == NULL)
97			continue;
98
99		low = *basep;
100		high = low + (*sizep);
101
102		for (i = 0; i < image->nr_segments; i++) {
103			begin = image->segment[i].mem;
104			end = begin + image->segment[i].memsz;
105
106			if ((begin < high) && (end > low))
107				return -ETXTBSY;
108		}
109	}
110
111	return 0;
112}
113#endif /* !CONFIG_PPC_BOOK3E */
114
115static void copy_segments(unsigned long ind)
116{
117	unsigned long entry;
118	unsigned long *ptr;
119	void *dest;
120	void *addr;
121
122	/*
123	 * We rely on kexec_load to create a lists that properly
124	 * initializes these pointers before they are used.
125	 * We will still crash if the list is wrong, but at least
126	 * the compiler will be quiet.
127	 */
128	ptr = NULL;
129	dest = NULL;
130
131	for (entry = ind; !(entry & IND_DONE); entry = *ptr++) {
132		addr = __va(entry & PAGE_MASK);
133
134		switch (entry & IND_FLAGS) {
135		case IND_DESTINATION:
136			dest = addr;
137			break;
138		case IND_INDIRECTION:
139			ptr = addr;
140			break;
141		case IND_SOURCE:
142			copy_page(dest, addr);
143			dest += PAGE_SIZE;
144		}
145	}
146}
147
148void kexec_copy_flush(struct kimage *image)
149{
150	long i, nr_segments = image->nr_segments;
151	struct  kexec_segment ranges[KEXEC_SEGMENT_MAX];
152
153	/* save the ranges on the stack to efficiently flush the icache */
154	memcpy(ranges, image->segment, sizeof(ranges));
155
156	/*
157	 * After this call we may not use anything allocated in dynamic
158	 * memory, including *image.
159	 *
160	 * Only globals and the stack are allowed.
161	 */
162	copy_segments(image->head);
163
164	/*
165	 * we need to clear the icache for all dest pages sometime,
166	 * including ones that were in place on the original copy
167	 */
168	for (i = 0; i < nr_segments; i++)
169		flush_icache_range((unsigned long)__va(ranges[i].mem),
170			(unsigned long)__va(ranges[i].mem + ranges[i].memsz));
171}
172
173#ifdef CONFIG_SMP
174
175static int kexec_all_irq_disabled = 0;
176
177static void kexec_smp_down(void *arg)
178{
179	local_irq_disable();
180	hard_irq_disable();
181
182	mb(); /* make sure our irqs are disabled before we say they are */
183	get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
184	while(kexec_all_irq_disabled == 0)
185		cpu_relax();
186	mb(); /* make sure all irqs are disabled before this */
187	hw_breakpoint_disable();
188	/*
189	 * Now every CPU has IRQs off, we can clear out any pending
190	 * IPIs and be sure that no more will come in after this.
191	 */
192	if (ppc_md.kexec_cpu_down)
193		ppc_md.kexec_cpu_down(0, 1);
194
195	kexec_smp_wait();
196	/* NOTREACHED */
197}
198
199static void kexec_prepare_cpus_wait(int wait_state)
200{
201	int my_cpu, i, notified=-1;
202
203	hw_breakpoint_disable();
204	my_cpu = get_cpu();
205	/* Make sure each CPU has at least made it to the state we need.
206	 *
207	 * FIXME: There is a (slim) chance of a problem if not all of the CPUs
208	 * are correctly onlined.  If somehow we start a CPU on boot with RTAS
209	 * start-cpu, but somehow that CPU doesn't write callin_cpu_map[] in
210	 * time, the boot CPU will timeout.  If it does eventually execute
211	 * stuff, the secondary will start up (paca[].cpu_start was written) and
212	 * get into a peculiar state.  If the platform supports
213	 * smp_ops->take_timebase(), the secondary CPU will probably be spinning
214	 * in there.  If not (i.e. pseries), the secondary will continue on and
215	 * try to online itself/idle/etc. If it survives that, we need to find
216	 * these possible-but-not-online-but-should-be CPUs and chaperone them
217	 * into kexec_smp_wait().
218	 */
219	for_each_online_cpu(i) {
220		if (i == my_cpu)
221			continue;
222
223		while (paca[i].kexec_state < wait_state) {
224			barrier();
225			if (i != notified) {
226				printk(KERN_INFO "kexec: waiting for cpu %d "
227				       "(physical %d) to enter %i state\n",
228				       i, paca[i].hw_cpu_id, wait_state);
229				notified = i;
230			}
231		}
232	}
233	mb();
234}
235
236/*
237 * We need to make sure each present CPU is online.  The next kernel will scan
238 * the device tree and assume primary threads are online and query secondary
239 * threads via RTAS to online them if required.  If we don't online primary
240 * threads, they will be stuck.  However, we also online secondary threads as we
241 * may be using 'cede offline'.  In this case RTAS doesn't see the secondary
242 * threads as offline -- and again, these CPUs will be stuck.
243 *
244 * So, we online all CPUs that should be running, including secondary threads.
245 */
246static void wake_offline_cpus(void)
247{
248	int cpu = 0;
249
250	for_each_present_cpu(cpu) {
251		if (!cpu_online(cpu)) {
252			printk(KERN_INFO "kexec: Waking offline cpu %d.\n",
253			       cpu);
254			WARN_ON(cpu_up(cpu));
255		}
256	}
257}
258
259static void kexec_prepare_cpus(void)
260{
261	wake_offline_cpus();
262	smp_call_function(kexec_smp_down, NULL, /* wait */0);
263	local_irq_disable();
264	hard_irq_disable();
265
266	mb(); /* make sure IRQs are disabled before we say they are */
267	get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
268
269	kexec_prepare_cpus_wait(KEXEC_STATE_IRQS_OFF);
270	/* we are sure every CPU has IRQs off at this point */
271	kexec_all_irq_disabled = 1;
272
273	/* after we tell the others to go down */
274	if (ppc_md.kexec_cpu_down)
275		ppc_md.kexec_cpu_down(0, 0);
276
277	/*
278	 * Before removing MMU mappings make sure all CPUs have entered real
279	 * mode:
280	 */
281	kexec_prepare_cpus_wait(KEXEC_STATE_REAL_MODE);
282
283	put_cpu();
284}
285
286#else /* ! SMP */
287
288static void kexec_prepare_cpus(void)
289{
290	/*
291	 * move the secondarys to us so that we can copy
292	 * the new kernel 0-0x100 safely
293	 *
294	 * do this if kexec in setup.c ?
295	 *
296	 * We need to release the cpus if we are ever going from an
297	 * UP to an SMP kernel.
298	 */
299	smp_release_cpus();
300	if (ppc_md.kexec_cpu_down)
301		ppc_md.kexec_cpu_down(0, 0);
302	local_irq_disable();
303	hard_irq_disable();
304}
305
306#endif /* SMP */
307
308/*
309 * kexec thread structure and stack.
310 *
311 * We need to make sure that this is 16384-byte aligned due to the
312 * way process stacks are handled.  It also must be statically allocated
313 * or allocated as part of the kimage, because everything else may be
314 * overwritten when we copy the kexec image.  We piggyback on the
315 * "init_task" linker section here to statically allocate a stack.
316 *
317 * We could use a smaller stack if we don't care about anything using
318 * current, but that audit has not been performed.
319 */
320static union thread_union kexec_stack __init_task_data =
321	{ };
322
323/*
324 * For similar reasons to the stack above, the kexecing CPU needs to be on a
325 * static PACA; we switch to kexec_paca.
326 */
327struct paca_struct kexec_paca;
328
329/* Our assembly helper, in misc_64.S */
330extern void kexec_sequence(void *newstack, unsigned long start,
331			   void *image, void *control,
332			   void (*clear_all)(void)) __noreturn;
333
334/* too late to fail here */
335void default_machine_kexec(struct kimage *image)
336{
337	/* prepare control code if any */
338
339	/*
340        * If the kexec boot is the normal one, need to shutdown other cpus
341        * into our wait loop and quiesce interrupts.
342        * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
343        * stopping other CPUs and collecting their pt_regs is done before
344        * using debugger IPI.
345        */
346
347	if (!kdump_in_progress())
348		kexec_prepare_cpus();
349
350	pr_debug("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	kexec_stack.thread_info.task = current_thread_info()->task;
357	kexec_stack.thread_info.flags = 0;
358	kexec_stack.thread_info.preempt_count = HARDIRQ_OFFSET;
359	kexec_stack.thread_info.cpu = current_thread_info()->cpu;
360
361	/* We need a static PACA, too; copy this CPU's PACA over and switch to
362	 * it.  Also poison per_cpu_offset to catch anyone using non-static
363	 * data.
364	 */
365	memcpy(&kexec_paca, get_paca(), sizeof(struct paca_struct));
366	kexec_paca.data_offset = 0xedeaddeadeeeeeeeUL;
367	paca = (struct paca_struct *)RELOC_HIDE(&kexec_paca, 0) -
368		kexec_paca.paca_index;
369	setup_paca(&kexec_paca);
370
371	/* XXX: If anyone does 'dynamic lppacas' this will also need to be
372	 * switched to a static version!
373	 */
374
375	/* Some things are best done in assembly.  Finding globals with
376	 * a toc is easier in C, so pass in what we can.
377	 */
378	kexec_sequence(&kexec_stack, image->start, image,
379			page_address(image->control_code_page),
380			ppc_md.hpte_clear_all);
381	/* NOTREACHED */
382}
383
384#ifndef CONFIG_PPC_BOOK3E
385/* Values we need to export to the second kernel via the device tree. */
386static unsigned long htab_base;
387static unsigned long htab_size;
388
389static struct property htab_base_prop = {
390	.name = "linux,htab-base",
391	.length = sizeof(unsigned long),
392	.value = &htab_base,
393};
394
395static struct property htab_size_prop = {
396	.name = "linux,htab-size",
397	.length = sizeof(unsigned long),
398	.value = &htab_size,
399};
400
401static int __init export_htab_values(void)
402{
403	struct device_node *node;
404	struct property *prop;
405
406	/* On machines with no htab htab_address is NULL */
407	if (!htab_address)
408		return -ENODEV;
409
410	node = of_find_node_by_path("/chosen");
411	if (!node)
412		return -ENODEV;
413
414	/* remove any stale propertys so ours can be found */
415	prop = of_find_property(node, htab_base_prop.name, NULL);
416	if (prop)
417		of_remove_property(node, prop);
418	prop = of_find_property(node, htab_size_prop.name, NULL);
419	if (prop)
420		of_remove_property(node, prop);
421
422	htab_base = cpu_to_be64(__pa(htab_address));
423	of_add_property(node, &htab_base_prop);
424	htab_size = cpu_to_be64(htab_size_bytes);
425	of_add_property(node, &htab_size_prop);
426
427	of_node_put(node);
428	return 0;
429}
430late_initcall(export_htab_values);
431#endif /* !CONFIG_PPC_BOOK3E */
432