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>
27 #include <linux/irq.h>
29 #include <asm/processor.h>
30 #include <asm/machdep.h>
31 #include <asm/kexec.h>
32 #include <asm/kdump.h>
34 #include <asm/firmware.h>
39 #define DBG(fmt...) udbg_printf(fmt)
44 /* This keeps a track of which one is crashing cpu. */
45 int crashing_cpu = -1;
46 static cpumask_t cpus_in_crash = CPU_MASK_NONE;
47 cpumask_t cpus_in_sr = CPU_MASK_NONE;
50 static atomic_t enter_on_soft_reset = ATOMIC_INIT(0);
52 void crash_ipi_callback(struct pt_regs *regs)
54 int cpu = smp_processor_id();
60 if (!cpu_isset(cpu, cpus_in_crash))
61 crash_save_cpu(regs, cpu);
62 cpu_set(cpu, cpus_in_crash);
65 * Entered via soft-reset - could be the kdump
66 * process is invoked using soft-reset or user activated
67 * it if some CPU did not respond to an IPI.
68 * For soft-reset, the secondary CPU can enter this func
69 * twice. 1 - using IPI, and 2. soft-reset.
70 * Tell the kexec CPU that entered via soft-reset and ready
73 if (cpu_isset(cpu, cpus_in_sr)) {
74 cpu_clear(cpu, cpus_in_sr);
75 atomic_inc(&enter_on_soft_reset);
79 * Starting the kdump boot.
80 * This barrier is needed to make sure that all CPUs are stopped.
81 * If not, soft-reset will be invoked to bring other CPUs.
83 while (!cpu_isset(crashing_cpu, cpus_in_crash))
86 if (ppc_md.kexec_cpu_down)
87 ppc_md.kexec_cpu_down(1, 1);
99 * Wait until all CPUs are entered via soft-reset.
101 static void crash_soft_reset_check(int cpu)
103 unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
105 cpu_clear(cpu, cpus_in_sr);
106 while (atomic_read(&enter_on_soft_reset) != ncpus)
111 static void crash_kexec_prepare_cpus(int cpu)
115 unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
117 crash_send_ipi(crash_ipi_callback);
121 * FIXME: Until we will have the way to stop other CPUSs reliabally,
122 * the crash CPU will send an IPI and wait for other CPUs to
124 * Delay of at least 10 seconds.
126 printk(KERN_EMERG "Sending IPI to other cpus...\n");
128 while ((cpus_weight(cpus_in_crash) < ncpus) && (--msecs > 0)) {
133 /* Would it be better to replace the trap vector here? */
136 * FIXME: In case if we do not get all CPUs, one possibility: ask the
137 * user to do soft reset such that we get all.
138 * Soft-reset will be used until better mechanism is implemented.
140 if (cpus_weight(cpus_in_crash) < ncpus) {
141 printk(KERN_EMERG "done waiting: %d cpu(s) not responding\n",
142 ncpus - cpus_weight(cpus_in_crash));
143 printk(KERN_EMERG "Activate soft-reset to stop other cpu(s)\n");
144 cpus_in_sr = CPU_MASK_NONE;
145 atomic_set(&enter_on_soft_reset, 0);
146 while (cpus_weight(cpus_in_crash) < ncpus)
150 * Make sure all CPUs are entered via soft-reset if the kdump is
151 * invoked using soft-reset.
153 if (cpu_isset(cpu, cpus_in_sr))
154 crash_soft_reset_check(cpu);
155 /* Leave the IPI callback set */
159 * This function will be called by secondary cpus or by kexec cpu
160 * if soft-reset is activated to stop some CPUs.
162 void crash_kexec_secondary(struct pt_regs *regs)
164 int cpu = smp_processor_id();
168 local_irq_save(flags);
169 /* Wait 5ms if the kexec CPU is not entered yet. */
170 while (crashing_cpu < 0) {
173 * Either kdump image is not loaded or
174 * kdump process is not started - Probably xmon
175 * exited using 'x'(exit and recover) or
176 * kexec_should_crash() failed for all running tasks.
178 cpu_clear(cpu, cpus_in_sr);
179 local_irq_restore(flags);
185 if (cpu == crashing_cpu) {
187 * Panic CPU will enter this func only via soft-reset.
188 * Wait until all secondary CPUs entered and
189 * then start kexec boot.
191 crash_soft_reset_check(cpu);
192 cpu_set(crashing_cpu, cpus_in_crash);
193 if (ppc_md.kexec_cpu_down)
194 ppc_md.kexec_cpu_down(1, 0);
195 machine_kexec(kexec_crash_image);
198 crash_ipi_callback(regs);
202 static void crash_kexec_prepare_cpus(int cpu)
205 * move the secondarys to us so that we can copy
206 * the new kernel 0-0x100 safely
208 * do this if kexec in setup.c ?
217 void crash_kexec_secondary(struct pt_regs *regs)
219 cpus_in_sr = CPU_MASK_NONE;
222 #ifdef CONFIG_SPU_BASE
225 #include <asm/spu_priv1.h>
227 struct crash_spu_info {
229 u32 saved_spu_runcntl_RW;
230 u32 saved_spu_status_R;
231 u32 saved_spu_npc_RW;
232 u64 saved_mfc_sr1_RW;
237 #define CRASH_NUM_SPUS 16 /* Enough for current hardware */
238 static struct crash_spu_info crash_spu_info[CRASH_NUM_SPUS];
240 static void crash_kexec_stop_spus(void)
246 for (i = 0; i < CRASH_NUM_SPUS; i++) {
247 if (!crash_spu_info[i].spu)
250 spu = crash_spu_info[i].spu;
252 crash_spu_info[i].saved_spu_runcntl_RW =
253 in_be32(&spu->problem->spu_runcntl_RW);
254 crash_spu_info[i].saved_spu_status_R =
255 in_be32(&spu->problem->spu_status_R);
256 crash_spu_info[i].saved_spu_npc_RW =
257 in_be32(&spu->problem->spu_npc_RW);
259 crash_spu_info[i].saved_mfc_dar = spu_mfc_dar_get(spu);
260 crash_spu_info[i].saved_mfc_dsisr = spu_mfc_dsisr_get(spu);
261 tmp = spu_mfc_sr1_get(spu);
262 crash_spu_info[i].saved_mfc_sr1_RW = tmp;
264 tmp &= ~MFC_STATE1_MASTER_RUN_CONTROL_MASK;
265 spu_mfc_sr1_set(spu, tmp);
271 void crash_register_spus(struct list_head *list)
275 list_for_each_entry(spu, list, full_list) {
276 if (WARN_ON(spu->number >= CRASH_NUM_SPUS))
279 crash_spu_info[spu->number].spu = spu;
284 static inline void crash_kexec_stop_spus(void)
287 #endif /* CONFIG_SPU_BASE */
289 void default_machine_crash_shutdown(struct pt_regs *regs)
294 * This function is only called after the system
295 * has panicked or is otherwise in a critical state.
296 * The minimum amount of code to allow a kexec'd kernel
297 * to run successfully needs to happen here.
299 * In practice this means stopping other cpus in
301 * The kernel is broken so disable interrupts.
306 struct irq_desc *desc = irq_desc + irq;
308 if (desc->status & IRQ_INPROGRESS)
309 desc->chip->eoi(irq);
311 if (!(desc->status & IRQ_DISABLED))
312 desc->chip->disable(irq);
316 * Make a note of crashing cpu. Will be used in machine_kexec
317 * such that another IPI will not be sent.
319 crashing_cpu = smp_processor_id();
320 crash_save_cpu(regs, crashing_cpu);
321 crash_kexec_prepare_cpus(crashing_cpu);
322 cpu_set(crashing_cpu, cpus_in_crash);
323 crash_kexec_stop_spus();
324 if (ppc_md.kexec_cpu_down)
325 ppc_md.kexec_cpu_down(1, 0);