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>
18 #include <linux/kernel.h>
19 #include <linux/cpu.h>
22 #include <asm/current.h>
23 #include <asm/machdep.h>
24 #include <asm/cacheflush.h>
27 #include <asm/sections.h> /* _end */
30 #include <asm/hw_breakpoint.h>
32 int default_machine_kexec_prepare(struct kimage
*image
)
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
;
41 if (!ppc_md
.hpte_clear_all
)
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.
49 for (i
= 0; i
< image
->nr_segments
; i
++)
50 if (image
->segment
[i
].mem
< __pa(_end
))
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.
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).
64 low
= __pa(htab_address
);
65 high
= low
+ htab_size_bytes
;
67 for (i
= 0; i
< image
->nr_segments
; i
++) {
68 begin
= image
->segment
[i
].mem
;
69 end
= begin
+ image
->segment
[i
].memsz
;
71 if ((begin
< high
) && (end
> low
))
76 /* We also should not overwrite the tce tables */
77 for (node
= of_find_node_by_type(NULL
, "pci"); node
!= NULL
;
78 node
= of_find_node_by_type(node
, "pci")) {
79 basep
= of_get_property(node
, "linux,tce-base", NULL
);
80 sizep
= of_get_property(node
, "linux,tce-size", NULL
);
81 if (basep
== NULL
|| sizep
== NULL
)
85 high
= low
+ (*sizep
);
87 for (i
= 0; i
< image
->nr_segments
; i
++) {
88 begin
= image
->segment
[i
].mem
;
89 end
= begin
+ image
->segment
[i
].memsz
;
91 if ((begin
< high
) && (end
> low
))
99 #define IND_FLAGS (IND_DESTINATION | IND_INDIRECTION | IND_DONE | IND_SOURCE)
101 static void copy_segments(unsigned long ind
)
109 * We rely on kexec_load to create a lists that properly
110 * initializes these pointers before they are used.
111 * We will still crash if the list is wrong, but at least
112 * the compiler will be quiet.
117 for (entry
= ind
; !(entry
& IND_DONE
); entry
= *ptr
++) {
118 addr
= __va(entry
& PAGE_MASK
);
120 switch (entry
& IND_FLAGS
) {
121 case IND_DESTINATION
:
124 case IND_INDIRECTION
:
128 copy_page(dest
, addr
);
134 void kexec_copy_flush(struct kimage
*image
)
136 long i
, nr_segments
= image
->nr_segments
;
137 struct kexec_segment ranges
[KEXEC_SEGMENT_MAX
];
139 /* save the ranges on the stack to efficiently flush the icache */
140 memcpy(ranges
, image
->segment
, sizeof(ranges
));
143 * After this call we may not use anything allocated in dynamic
144 * memory, including *image.
146 * Only globals and the stack are allowed.
148 copy_segments(image
->head
);
151 * we need to clear the icache for all dest pages sometime,
152 * including ones that were in place on the original copy
154 for (i
= 0; i
< nr_segments
; i
++)
155 flush_icache_range((unsigned long)__va(ranges
[i
].mem
),
156 (unsigned long)__va(ranges
[i
].mem
+ ranges
[i
].memsz
));
161 static int kexec_all_irq_disabled
= 0;
163 static void kexec_smp_down(void *arg
)
166 mb(); /* make sure our irqs are disabled before we say they are */
167 get_paca()->kexec_state
= KEXEC_STATE_IRQS_OFF
;
168 while(kexec_all_irq_disabled
== 0)
170 mb(); /* make sure all irqs are disabled before this */
171 hw_breakpoint_disable();
173 * Now every CPU has IRQs off, we can clear out any pending
174 * IPIs and be sure that no more will come in after this.
176 if (ppc_md
.kexec_cpu_down
)
177 ppc_md
.kexec_cpu_down(0, 1);
183 static void kexec_prepare_cpus_wait(int wait_state
)
185 int my_cpu
, i
, notified
=-1;
187 hw_breakpoint_disable();
189 for_each_online_cpu(i
) {
193 while (paca
[i
].kexec_state
< wait_state
) {
196 printk(KERN_INFO
"kexec: waiting for cpu %d "
197 "(physical %d) to enter %i state\n",
198 i
, paca
[i
].hw_cpu_id
, wait_state
);
207 * We need to make sure each present CPU is online. The next kernel will scan
208 * the device tree and assume primary threads are online and query secondary
209 * threads via RTAS to online them if required. If we don't online primary
210 * threads, they will be stuck. However, we also online secondary threads as we
211 * may be using 'cede offline'. In this case RTAS doesn't see the secondary
212 * threads as offline -- and again, these CPUs will be stuck.
214 * So, we online all CPUs that should be running, including secondary threads.
216 static void wake_offline_cpus(void)
220 for_each_present_cpu(cpu
) {
221 if (!cpu_online(cpu
)) {
222 printk(KERN_INFO
"kexec: Waking offline cpu %d.\n",
229 static void kexec_prepare_cpus(void)
232 smp_call_function(kexec_smp_down
, NULL
, /* wait */0);
234 mb(); /* make sure IRQs are disabled before we say they are */
235 get_paca()->kexec_state
= KEXEC_STATE_IRQS_OFF
;
237 kexec_prepare_cpus_wait(KEXEC_STATE_IRQS_OFF
);
238 /* we are sure every CPU has IRQs off at this point */
239 kexec_all_irq_disabled
= 1;
241 /* after we tell the others to go down */
242 if (ppc_md
.kexec_cpu_down
)
243 ppc_md
.kexec_cpu_down(0, 0);
246 * Before removing MMU mappings make sure all CPUs have entered real
249 kexec_prepare_cpus_wait(KEXEC_STATE_REAL_MODE
);
256 static void kexec_prepare_cpus(void)
259 * move the secondarys to us so that we can copy
260 * the new kernel 0-0x100 safely
262 * do this if kexec in setup.c ?
264 * We need to release the cpus if we are ever going from an
265 * UP to an SMP kernel.
268 if (ppc_md
.kexec_cpu_down
)
269 ppc_md
.kexec_cpu_down(0, 0);
276 * kexec thread structure and stack.
278 * We need to make sure that this is 16384-byte aligned due to the
279 * way process stacks are handled. It also must be statically allocated
280 * or allocated as part of the kimage, because everything else may be
281 * overwritten when we copy the kexec image. We piggyback on the
282 * "init_task" linker section here to statically allocate a stack.
284 * We could use a smaller stack if we don't care about anything using
285 * current, but that audit has not been performed.
287 static union thread_union kexec_stack __init_task_data
=
291 * For similar reasons to the stack above, the kexecing CPU needs to be on a
292 * static PACA; we switch to kexec_paca.
294 struct paca_struct kexec_paca
;
296 /* Our assembly helper, in kexec_stub.S */
297 extern NORET_TYPE
void kexec_sequence(void *newstack
, unsigned long start
,
298 void *image
, void *control
,
299 void (*clear_all
)(void)) ATTRIB_NORET
;
301 /* too late to fail here */
302 void default_machine_kexec(struct kimage
*image
)
304 /* prepare control code if any */
307 * If the kexec boot is the normal one, need to shutdown other cpus
308 * into our wait loop and quiesce interrupts.
309 * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
310 * stopping other CPUs and collecting their pt_regs is done before
311 * using debugger IPI.
314 if (crashing_cpu
== -1)
315 kexec_prepare_cpus();
317 pr_debug("kexec: Starting switchover sequence.\n");
319 kexec_stack
.thread_info
.task
= current_thread_info()->task
;
320 kexec_stack
.thread_info
.flags
= 0;
322 /* We need a static PACA, too; copy this CPU's PACA over and switch to
323 * it. Also poison per_cpu_offset to catch anyone using non-static
326 memcpy(&kexec_paca
, get_paca(), sizeof(struct paca_struct
));
327 kexec_paca
.data_offset
= 0xedeaddeadeeeeeeeUL
;
328 paca
= (struct paca_struct
*)RELOC_HIDE(&kexec_paca
, 0) -
329 kexec_paca
.paca_index
;
330 setup_paca(&kexec_paca
);
333 /* Some things are best done in assembly. Finding globals with
334 * a toc is easier in C, so pass in what we can.
336 kexec_sequence(&kexec_stack
, image
->start
, image
,
337 page_address(image
->control_code_page
),
338 ppc_md
.hpte_clear_all
);
342 /* Values we need to export to the second kernel via the device tree. */
343 static unsigned long htab_base
;
345 static struct property htab_base_prop
= {
346 .name
= "linux,htab-base",
347 .length
= sizeof(unsigned long),
351 static struct property htab_size_prop
= {
352 .name
= "linux,htab-size",
353 .length
= sizeof(unsigned long),
354 .value
= &htab_size_bytes
,
357 static int __init
export_htab_values(void)
359 struct device_node
*node
;
360 struct property
*prop
;
362 /* On machines with no htab htab_address is NULL */
366 node
= of_find_node_by_path("/chosen");
370 /* remove any stale propertys so ours can be found */
371 prop
= of_find_property(node
, htab_base_prop
.name
, NULL
);
373 prom_remove_property(node
, prop
);
374 prop
= of_find_property(node
, htab_size_prop
.name
, NULL
);
376 prom_remove_property(node
, prop
);
378 htab_base
= __pa(htab_address
);
379 prom_add_property(node
, &htab_base_prop
);
380 prom_add_property(node
, &htab_size_prop
);
385 late_initcall(export_htab_values
);