2 * Architecture specific (PPC64) functions for kexec based crash dumps.
4 * Copyright (C) 2005, IBM Corp.
6 * Created by: Haren Myneni
8 * This source code is licensed under the GNU General Public License,
9 * Version 2. See the file COPYING for more details.
15 #include <linux/kernel.h>
16 #include <linux/smp.h>
17 #include <linux/reboot.h>
18 #include <linux/kexec.h>
19 #include <linux/bootmem.h>
20 #include <linux/crash_dump.h>
21 #include <linux/delay.h>
22 #include <linux/elf.h>
23 #include <linux/elfcore.h>
24 #include <linux/init.h>
25 #include <linux/irq.h>
26 #include <linux/types.h>
28 #include <asm/processor.h>
29 #include <asm/machdep.h>
30 #include <asm/kexec.h>
31 #include <asm/kdump.h>
33 #include <asm/firmware.h>
38 #define DBG(fmt...) udbg_printf(fmt)
43 /* This keeps a track of which one is crashing cpu. */
44 int crashing_cpu = -1;
45 static cpumask_t cpus_in_crash = CPU_MASK_NONE;
46 cpumask_t cpus_in_sr = CPU_MASK_NONE;
49 static atomic_t enter_on_soft_reset = ATOMIC_INIT(0);
51 void crash_ipi_callback(struct pt_regs *regs)
53 int cpu = smp_processor_id();
59 if (!cpu_isset(cpu, cpus_in_crash))
60 crash_save_cpu(regs, cpu);
61 cpu_set(cpu, cpus_in_crash);
64 * Entered via soft-reset - could be the kdump
65 * process is invoked using soft-reset or user activated
66 * it if some CPU did not respond to an IPI.
67 * For soft-reset, the secondary CPU can enter this func
68 * twice. 1 - using IPI, and 2. soft-reset.
69 * Tell the kexec CPU that entered via soft-reset and ready
72 if (cpu_isset(cpu, cpus_in_sr)) {
73 cpu_clear(cpu, cpus_in_sr);
74 atomic_inc(&enter_on_soft_reset);
78 * Starting the kdump boot.
79 * This barrier is needed to make sure that all CPUs are stopped.
80 * If not, soft-reset will be invoked to bring other CPUs.
82 while (!cpu_isset(crashing_cpu, cpus_in_crash))
85 if (ppc_md.kexec_cpu_down)
86 ppc_md.kexec_cpu_down(1, 1);
98 * Wait until all CPUs are entered via soft-reset.
100 static void crash_soft_reset_check(int cpu)
102 unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
104 cpu_clear(cpu, cpus_in_sr);
105 while (atomic_read(&enter_on_soft_reset) != ncpus)
110 static void crash_kexec_prepare_cpus(int cpu)
114 unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
116 crash_send_ipi(crash_ipi_callback);
120 * FIXME: Until we will have the way to stop other CPUSs reliabally,
121 * the crash CPU will send an IPI and wait for other CPUs to
123 * Delay of at least 10 seconds.
125 printk(KERN_EMERG "Sending IPI to other cpus...\n");
127 while ((cpus_weight(cpus_in_crash) < ncpus) && (--msecs > 0)) {
132 /* Would it be better to replace the trap vector here? */
135 * FIXME: In case if we do not get all CPUs, one possibility: ask the
136 * user to do soft reset such that we get all.
137 * Soft-reset will be used until better mechanism is implemented.
139 if (cpus_weight(cpus_in_crash) < ncpus) {
140 printk(KERN_EMERG "done waiting: %d cpu(s) not responding\n",
141 ncpus - cpus_weight(cpus_in_crash));
142 printk(KERN_EMERG "Activate soft-reset to stop other cpu(s)\n");
143 cpus_in_sr = CPU_MASK_NONE;
144 atomic_set(&enter_on_soft_reset, 0);
145 while (cpus_weight(cpus_in_crash) < ncpus)
149 * Make sure all CPUs are entered via soft-reset if the kdump is
150 * invoked using soft-reset.
152 if (cpu_isset(cpu, cpus_in_sr))
153 crash_soft_reset_check(cpu);
154 /* Leave the IPI callback set */
158 * This function will be called by secondary cpus or by kexec cpu
159 * if soft-reset is activated to stop some CPUs.
161 void crash_kexec_secondary(struct pt_regs *regs)
163 int cpu = smp_processor_id();
167 local_irq_save(flags);
168 /* Wait 5ms if the kexec CPU is not entered yet. */
169 while (crashing_cpu < 0) {
172 * Either kdump image is not loaded or
173 * kdump process is not started - Probably xmon
174 * exited using 'x'(exit and recover) or
175 * kexec_should_crash() failed for all running tasks.
177 cpu_clear(cpu, cpus_in_sr);
178 local_irq_restore(flags);
184 if (cpu == crashing_cpu) {
186 * Panic CPU will enter this func only via soft-reset.
187 * Wait until all secondary CPUs entered and
188 * then start kexec boot.
190 crash_soft_reset_check(cpu);
191 cpu_set(crashing_cpu, cpus_in_crash);
192 if (ppc_md.kexec_cpu_down)
193 ppc_md.kexec_cpu_down(1, 0);
194 machine_kexec(kexec_crash_image);
197 crash_ipi_callback(regs);
201 static void crash_kexec_prepare_cpus(int cpu)
204 * move the secondarys to us so that we can copy
205 * the new kernel 0-0x100 safely
207 * do this if kexec in setup.c ?
216 void crash_kexec_secondary(struct pt_regs *regs)
218 cpus_in_sr = CPU_MASK_NONE;
221 #ifdef CONFIG_SPU_BASE
224 #include <asm/spu_priv1.h>
226 struct crash_spu_info {
228 u32 saved_spu_runcntl_RW;
229 u32 saved_spu_status_R;
230 u32 saved_spu_npc_RW;
231 u64 saved_mfc_sr1_RW;
236 #define CRASH_NUM_SPUS 16 /* Enough for current hardware */
237 static struct crash_spu_info crash_spu_info[CRASH_NUM_SPUS];
239 static void crash_kexec_stop_spus(void)
245 for (i = 0; i < CRASH_NUM_SPUS; i++) {
246 if (!crash_spu_info[i].spu)
249 spu = crash_spu_info[i].spu;
251 crash_spu_info[i].saved_spu_runcntl_RW =
252 in_be32(&spu->problem->spu_runcntl_RW);
253 crash_spu_info[i].saved_spu_status_R =
254 in_be32(&spu->problem->spu_status_R);
255 crash_spu_info[i].saved_spu_npc_RW =
256 in_be32(&spu->problem->spu_npc_RW);
258 crash_spu_info[i].saved_mfc_dar = spu_mfc_dar_get(spu);
259 crash_spu_info[i].saved_mfc_dsisr = spu_mfc_dsisr_get(spu);
260 tmp = spu_mfc_sr1_get(spu);
261 crash_spu_info[i].saved_mfc_sr1_RW = tmp;
263 tmp &= ~MFC_STATE1_MASTER_RUN_CONTROL_MASK;
264 spu_mfc_sr1_set(spu, tmp);
270 void crash_register_spus(struct list_head *list)
274 list_for_each_entry(spu, list, full_list) {
275 if (WARN_ON(spu->number >= CRASH_NUM_SPUS))
278 crash_spu_info[spu->number].spu = spu;
283 static inline void crash_kexec_stop_spus(void)
286 #endif /* CONFIG_SPU_BASE */
288 void default_machine_crash_shutdown(struct pt_regs *regs)
293 * This function is only called after the system
294 * has panicked or is otherwise in a critical state.
295 * The minimum amount of code to allow a kexec'd kernel
296 * to run successfully needs to happen here.
298 * In practice this means stopping other cpus in
300 * The kernel is broken so disable interrupts.
305 struct irq_desc *desc = irq_desc + irq;
307 if (desc->status & IRQ_INPROGRESS)
308 desc->chip->eoi(irq);
310 if (!(desc->status & IRQ_DISABLED))
311 desc->chip->disable(irq);
315 * Make a note of crashing cpu. Will be used in machine_kexec
316 * such that another IPI will not be sent.
318 crashing_cpu = smp_processor_id();
319 crash_save_cpu(regs, crashing_cpu);
320 crash_kexec_prepare_cpus(crashing_cpu);
321 cpu_set(crashing_cpu, cpus_in_crash);
322 crash_kexec_stop_spus();
323 if (ppc_md.kexec_cpu_down)
324 ppc_md.kexec_cpu_down(1, 0);