2 * PPC64 code to handle Linux booting another kernel.
4 * Copyright (C) 2004-2005, IBM Corp.
6 * Created by: Milton D Miller II
8 * This source code is licensed under the GNU General Public License,
9 * Version 2. See the file COPYING for more details.
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>
20 #include <asm/current.h>
21 #include <asm/machdep.h>
22 #include <asm/cacheflush.h>
25 #include <asm/sections.h> /* _end */
29 int default_machine_kexec_prepare(struct kimage
*image
)
32 unsigned long begin
, end
; /* limits of segment */
33 unsigned long low
, high
; /* limits of blocked memory range */
34 struct device_node
*node
;
35 const unsigned long *basep
;
36 const unsigned int *sizep
;
38 if (!ppc_md
.hpte_clear_all
)
42 * Since we use the kernel fault handlers and paging code to
43 * handle the virtual mode, we must make sure no destination
44 * overlaps kernel static data or bss.
46 for (i
= 0; i
< image
->nr_segments
; i
++)
47 if (image
->segment
[i
].mem
< __pa(_end
))
51 * For non-LPAR, we absolutely can not overwrite the mmu hash
52 * table, since we are still using the bolted entries in it to
53 * do the copy. Check that here.
55 * It is safe if the end is below the start of the blocked
56 * region (end <= low), or if the beginning is after the
57 * end of the blocked region (begin >= high). Use the
58 * boolean identity !(a || b) === (!a && !b).
61 low
= __pa(htab_address
);
62 high
= low
+ htab_size_bytes
;
64 for (i
= 0; i
< image
->nr_segments
; i
++) {
65 begin
= image
->segment
[i
].mem
;
66 end
= begin
+ image
->segment
[i
].memsz
;
68 if ((begin
< high
) && (end
> low
))
73 /* We also should not overwrite the tce tables */
74 for (node
= of_find_node_by_type(NULL
, "pci"); node
!= NULL
;
75 node
= of_find_node_by_type(node
, "pci")) {
76 basep
= of_get_property(node
, "linux,tce-base", NULL
);
77 sizep
= of_get_property(node
, "linux,tce-size", NULL
);
78 if (basep
== NULL
|| sizep
== NULL
)
82 high
= low
+ (*sizep
);
84 for (i
= 0; i
< image
->nr_segments
; i
++) {
85 begin
= image
->segment
[i
].mem
;
86 end
= begin
+ image
->segment
[i
].memsz
;
88 if ((begin
< high
) && (end
> low
))
96 #define IND_FLAGS (IND_DESTINATION | IND_INDIRECTION | IND_DONE | IND_SOURCE)
98 static void copy_segments(unsigned long ind
)
106 * We rely on kexec_load to create a lists that properly
107 * initializes these pointers before they are used.
108 * We will still crash if the list is wrong, but at least
109 * the compiler will be quiet.
114 for (entry
= ind
; !(entry
& IND_DONE
); entry
= *ptr
++) {
115 addr
= __va(entry
& PAGE_MASK
);
117 switch (entry
& IND_FLAGS
) {
118 case IND_DESTINATION
:
121 case IND_INDIRECTION
:
125 copy_page(dest
, addr
);
131 void kexec_copy_flush(struct kimage
*image
)
133 long i
, nr_segments
= image
->nr_segments
;
134 struct kexec_segment ranges
[KEXEC_SEGMENT_MAX
];
136 /* save the ranges on the stack to efficiently flush the icache */
137 memcpy(ranges
, image
->segment
, sizeof(ranges
));
140 * After this call we may not use anything allocated in dynamic
141 * memory, including *image.
143 * Only globals and the stack are allowed.
145 copy_segments(image
->head
);
148 * we need to clear the icache for all dest pages sometime,
149 * including ones that were in place on the original copy
151 for (i
= 0; i
< nr_segments
; i
++)
152 flush_icache_range((unsigned long)__va(ranges
[i
].mem
),
153 (unsigned long)__va(ranges
[i
].mem
+ ranges
[i
].memsz
));
158 static int kexec_all_irq_disabled
= 0;
160 static void kexec_smp_down(void *arg
)
163 mb(); /* make sure our irqs are disabled before we say they are */
164 get_paca()->kexec_state
= KEXEC_STATE_IRQS_OFF
;
165 while(kexec_all_irq_disabled
== 0)
167 mb(); /* make sure all irqs are disabled before this */
169 * Now every CPU has IRQs off, we can clear out any pending
170 * IPIs and be sure that no more will come in after this.
172 if (ppc_md
.kexec_cpu_down
)
173 ppc_md
.kexec_cpu_down(0, 1);
179 static void kexec_prepare_cpus_wait(int wait_state
)
181 int my_cpu
, i
, notified
=-1;
184 /* Make sure each CPU has atleast made it to the state we need */
185 for_each_online_cpu(i
) {
189 while (paca
[i
].kexec_state
< wait_state
) {
192 printk( "kexec: waiting for cpu %d (physical"
193 " %d) to enter %i state\n",
194 i
, paca
[i
].hw_cpu_id
, wait_state
);
202 static void kexec_prepare_cpus(void)
205 smp_call_function(kexec_smp_down
, NULL
, /* wait */0);
207 mb(); /* make sure IRQs are disabled before we say they are */
208 get_paca()->kexec_state
= KEXEC_STATE_IRQS_OFF
;
210 kexec_prepare_cpus_wait(KEXEC_STATE_IRQS_OFF
);
211 /* we are sure every CPU has IRQs off at this point */
212 kexec_all_irq_disabled
= 1;
214 /* after we tell the others to go down */
215 if (ppc_md
.kexec_cpu_down
)
216 ppc_md
.kexec_cpu_down(0, 0);
218 /* Before removing MMU mapings make sure all CPUs have entered real mode */
219 kexec_prepare_cpus_wait(KEXEC_STATE_REAL_MODE
);
226 static void kexec_prepare_cpus(void)
229 * move the secondarys to us so that we can copy
230 * the new kernel 0-0x100 safely
232 * do this if kexec in setup.c ?
234 * We need to release the cpus if we are ever going from an
235 * UP to an SMP kernel.
238 if (ppc_md
.kexec_cpu_down
)
239 ppc_md
.kexec_cpu_down(0, 0);
246 * kexec thread structure and stack.
248 * We need to make sure that this is 16384-byte aligned due to the
249 * way process stacks are handled. It also must be statically allocated
250 * or allocated as part of the kimage, because everything else may be
251 * overwritten when we copy the kexec image. We piggyback on the
252 * "init_task" linker section here to statically allocate a stack.
254 * We could use a smaller stack if we don't care about anything using
255 * current, but that audit has not been performed.
257 static union thread_union kexec_stack __init_task_data
=
260 /* Our assembly helper, in kexec_stub.S */
261 extern NORET_TYPE
void kexec_sequence(void *newstack
, unsigned long start
,
262 void *image
, void *control
,
263 void (*clear_all
)(void)) ATTRIB_NORET
;
265 /* too late to fail here */
266 void default_machine_kexec(struct kimage
*image
)
268 /* prepare control code if any */
271 * If the kexec boot is the normal one, need to shutdown other cpus
272 * into our wait loop and quiesce interrupts.
273 * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
274 * stopping other CPUs and collecting their pt_regs is done before
275 * using debugger IPI.
278 if (crashing_cpu
== -1)
279 kexec_prepare_cpus();
281 /* switch to a staticly allocated stack. Based on irq stack code.
282 * XXX: the task struct will likely be invalid once we do the copy!
284 kexec_stack
.thread_info
.task
= current_thread_info()->task
;
285 kexec_stack
.thread_info
.flags
= 0;
287 /* Some things are best done in assembly. Finding globals with
288 * a toc is easier in C, so pass in what we can.
290 kexec_sequence(&kexec_stack
, image
->start
, image
,
291 page_address(image
->control_code_page
),
292 ppc_md
.hpte_clear_all
);
296 /* Values we need to export to the second kernel via the device tree. */
297 static unsigned long htab_base
;
299 static struct property htab_base_prop
= {
300 .name
= "linux,htab-base",
301 .length
= sizeof(unsigned long),
305 static struct property htab_size_prop
= {
306 .name
= "linux,htab-size",
307 .length
= sizeof(unsigned long),
308 .value
= &htab_size_bytes
,
311 static int __init
export_htab_values(void)
313 struct device_node
*node
;
314 struct property
*prop
;
316 /* On machines with no htab htab_address is NULL */
320 node
= of_find_node_by_path("/chosen");
324 /* remove any stale propertys so ours can be found */
325 prop
= of_find_property(node
, htab_base_prop
.name
, NULL
);
327 prom_remove_property(node
, prop
);
328 prop
= of_find_property(node
, htab_size_prop
.name
, NULL
);
330 prom_remove_property(node
, prop
);
332 htab_base
= __pa(htab_address
);
333 prom_add_property(node
, &htab_base_prop
);
334 prom_add_property(node
, &htab_size_prop
);
339 late_initcall(export_htab_values
);