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>
20 #include <linux/hardirq.h>
23 #include <asm/current.h>
24 #include <asm/machdep.h>
25 #include <asm/cacheflush.h>
28 #include <asm/sections.h> /* _end */
31 #include <asm/hw_breakpoint.h>
33 #ifdef CONFIG_PPC_BOOK3E
34 int default_machine_kexec_prepare(struct kimage
*image
)
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.
42 for (i
= 0; i
< image
->nr_segments
; i
++)
43 if (image
->segment
[i
].mem
< __pa(_end
))
48 int default_machine_kexec_prepare(struct kimage
*image
)
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
;
57 if (!ppc_md
.hpte_clear_all
)
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.
65 for (i
= 0; i
< image
->nr_segments
; i
++)
66 if (image
->segment
[i
].mem
< __pa(_end
))
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.
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).
79 #ifdef CONFIG_PPC_STD_MMU_64
81 low
= __pa(htab_address
);
82 high
= low
+ htab_size_bytes
;
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
))
92 #endif /* CONFIG_PPC_STD_MMU_64 */
94 /* We also should not overwrite the tce tables */
95 for_each_node_by_type(node
, "pci") {
96 basep
= of_get_property(node
, "linux,tce-base", NULL
);
97 sizep
= of_get_property(node
, "linux,tce-size", NULL
);
98 if (basep
== NULL
|| sizep
== NULL
)
102 high
= low
+ (*sizep
);
104 for (i
= 0; i
< image
->nr_segments
; i
++) {
105 begin
= image
->segment
[i
].mem
;
106 end
= begin
+ image
->segment
[i
].memsz
;
108 if ((begin
< high
) && (end
> low
))
115 #endif /* !CONFIG_PPC_BOOK3E */
117 static void copy_segments(unsigned long ind
)
125 * We rely on kexec_load to create a lists that properly
126 * initializes these pointers before they are used.
127 * We will still crash if the list is wrong, but at least
128 * the compiler will be quiet.
133 for (entry
= ind
; !(entry
& IND_DONE
); entry
= *ptr
++) {
134 addr
= __va(entry
& PAGE_MASK
);
136 switch (entry
& IND_FLAGS
) {
137 case IND_DESTINATION
:
140 case IND_INDIRECTION
:
144 copy_page(dest
, addr
);
150 void kexec_copy_flush(struct kimage
*image
)
152 long i
, nr_segments
= image
->nr_segments
;
153 struct kexec_segment ranges
[KEXEC_SEGMENT_MAX
];
155 /* save the ranges on the stack to efficiently flush the icache */
156 memcpy(ranges
, image
->segment
, sizeof(ranges
));
159 * After this call we may not use anything allocated in dynamic
160 * memory, including *image.
162 * Only globals and the stack are allowed.
164 copy_segments(image
->head
);
167 * we need to clear the icache for all dest pages sometime,
168 * including ones that were in place on the original copy
170 for (i
= 0; i
< nr_segments
; i
++)
171 flush_icache_range((unsigned long)__va(ranges
[i
].mem
),
172 (unsigned long)__va(ranges
[i
].mem
+ ranges
[i
].memsz
));
177 static int kexec_all_irq_disabled
= 0;
179 static void kexec_smp_down(void *arg
)
184 mb(); /* make sure our irqs are disabled before we say they are */
185 get_paca()->kexec_state
= KEXEC_STATE_IRQS_OFF
;
186 while(kexec_all_irq_disabled
== 0)
188 mb(); /* make sure all irqs are disabled before this */
189 hw_breakpoint_disable();
191 * Now every CPU has IRQs off, we can clear out any pending
192 * IPIs and be sure that no more will come in after this.
194 if (ppc_md
.kexec_cpu_down
)
195 ppc_md
.kexec_cpu_down(0, 1);
201 static void kexec_prepare_cpus_wait(int wait_state
)
203 int my_cpu
, i
, notified
=-1;
205 hw_breakpoint_disable();
207 /* Make sure each CPU has at least made it to the state we need.
209 * FIXME: There is a (slim) chance of a problem if not all of the CPUs
210 * are correctly onlined. If somehow we start a CPU on boot with RTAS
211 * start-cpu, but somehow that CPU doesn't write callin_cpu_map[] in
212 * time, the boot CPU will timeout. If it does eventually execute
213 * stuff, the secondary will start up (paca[].cpu_start was written) and
214 * get into a peculiar state. If the platform supports
215 * smp_ops->take_timebase(), the secondary CPU will probably be spinning
216 * in there. If not (i.e. pseries), the secondary will continue on and
217 * try to online itself/idle/etc. If it survives that, we need to find
218 * these possible-but-not-online-but-should-be CPUs and chaperone them
219 * into kexec_smp_wait().
221 for_each_online_cpu(i
) {
225 while (paca
[i
].kexec_state
< wait_state
) {
228 printk(KERN_INFO
"kexec: waiting for cpu %d "
229 "(physical %d) to enter %i state\n",
230 i
, paca
[i
].hw_cpu_id
, wait_state
);
239 * We need to make sure each present CPU is online. The next kernel will scan
240 * the device tree and assume primary threads are online and query secondary
241 * threads via RTAS to online them if required. If we don't online primary
242 * threads, they will be stuck. However, we also online secondary threads as we
243 * may be using 'cede offline'. In this case RTAS doesn't see the secondary
244 * threads as offline -- and again, these CPUs will be stuck.
246 * So, we online all CPUs that should be running, including secondary threads.
248 static void wake_offline_cpus(void)
252 for_each_present_cpu(cpu
) {
253 if (!cpu_online(cpu
)) {
254 printk(KERN_INFO
"kexec: Waking offline cpu %d.\n",
256 WARN_ON(cpu_up(cpu
));
261 static void kexec_prepare_cpus(void)
264 smp_call_function(kexec_smp_down
, NULL
, /* wait */0);
268 mb(); /* make sure IRQs are disabled before we say they are */
269 get_paca()->kexec_state
= KEXEC_STATE_IRQS_OFF
;
271 kexec_prepare_cpus_wait(KEXEC_STATE_IRQS_OFF
);
272 /* we are sure every CPU has IRQs off at this point */
273 kexec_all_irq_disabled
= 1;
275 /* after we tell the others to go down */
276 if (ppc_md
.kexec_cpu_down
)
277 ppc_md
.kexec_cpu_down(0, 0);
280 * Before removing MMU mappings make sure all CPUs have entered real
283 kexec_prepare_cpus_wait(KEXEC_STATE_REAL_MODE
);
290 static void kexec_prepare_cpus(void)
293 * move the secondarys to us so that we can copy
294 * the new kernel 0-0x100 safely
296 * do this if kexec in setup.c ?
298 * We need to release the cpus if we are ever going from an
299 * UP to an SMP kernel.
302 if (ppc_md
.kexec_cpu_down
)
303 ppc_md
.kexec_cpu_down(0, 0);
311 * kexec thread structure and stack.
313 * We need to make sure that this is 16384-byte aligned due to the
314 * way process stacks are handled. It also must be statically allocated
315 * or allocated as part of the kimage, because everything else may be
316 * overwritten when we copy the kexec image. We piggyback on the
317 * "init_task" linker section here to statically allocate a stack.
319 * We could use a smaller stack if we don't care about anything using
320 * current, but that audit has not been performed.
322 static union thread_union kexec_stack __init_task_data
=
326 * For similar reasons to the stack above, the kexecing CPU needs to be on a
327 * static PACA; we switch to kexec_paca.
329 struct paca_struct kexec_paca
;
331 /* Our assembly helper, in misc_64.S */
332 extern void kexec_sequence(void *newstack
, unsigned long start
,
333 void *image
, void *control
,
334 void (*clear_all
)(void)) __noreturn
;
336 /* too late to fail here */
337 void default_machine_kexec(struct kimage
*image
)
339 /* prepare control code if any */
342 * If the kexec boot is the normal one, need to shutdown other cpus
343 * into our wait loop and quiesce interrupts.
344 * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
345 * stopping other CPUs and collecting their pt_regs is done before
346 * using debugger IPI.
349 if (!kdump_in_progress())
350 kexec_prepare_cpus();
352 pr_debug("kexec: Starting switchover sequence.\n");
354 /* switch to a staticly allocated stack. Based on irq stack code.
355 * We setup preempt_count to avoid using VMX in memcpy.
356 * XXX: the task struct will likely be invalid once we do the copy!
358 kexec_stack
.thread_info
.task
= current_thread_info()->task
;
359 kexec_stack
.thread_info
.flags
= 0;
360 kexec_stack
.thread_info
.preempt_count
= HARDIRQ_OFFSET
;
361 kexec_stack
.thread_info
.cpu
= current_thread_info()->cpu
;
363 /* We need a static PACA, too; copy this CPU's PACA over and switch to
364 * it. Also poison per_cpu_offset to catch anyone using non-static
367 memcpy(&kexec_paca
, get_paca(), sizeof(struct paca_struct
));
368 kexec_paca
.data_offset
= 0xedeaddeadeeeeeeeUL
;
369 paca
= (struct paca_struct
*)RELOC_HIDE(&kexec_paca
, 0) -
370 kexec_paca
.paca_index
;
371 setup_paca(&kexec_paca
);
373 /* XXX: If anyone does 'dynamic lppacas' this will also need to be
374 * switched to a static version!
377 /* Some things are best done in assembly. Finding globals with
378 * a toc is easier in C, so pass in what we can.
380 kexec_sequence(&kexec_stack
, image
->start
, image
,
381 page_address(image
->control_code_page
),
382 ppc_md
.hpte_clear_all
);
386 #ifdef CONFIG_PPC_STD_MMU_64
387 /* Values we need to export to the second kernel via the device tree. */
388 static unsigned long htab_base
;
389 static unsigned long htab_size
;
391 static struct property htab_base_prop
= {
392 .name
= "linux,htab-base",
393 .length
= sizeof(unsigned long),
397 static struct property htab_size_prop
= {
398 .name
= "linux,htab-size",
399 .length
= sizeof(unsigned long),
403 static int __init
export_htab_values(void)
405 struct device_node
*node
;
407 /* On machines with no htab htab_address is NULL */
411 node
= of_find_node_by_path("/chosen");
415 /* remove any stale propertys so ours can be found */
416 of_remove_property(node
, of_find_property(node
, htab_base_prop
.name
, NULL
));
417 of_remove_property(node
, of_find_property(node
, htab_size_prop
.name
, NULL
));
419 htab_base
= cpu_to_be64(__pa(htab_address
));
420 of_add_property(node
, &htab_base_prop
);
421 htab_size
= cpu_to_be64(htab_size_bytes
);
422 of_add_property(node
, &htab_size_prop
);
427 late_initcall(export_htab_values
);
428 #endif /* CONFIG_PPC_STD_MMU_64 */