1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Architecture specific (PPC64) functions for kexec based crash dumps.
  4 *
  5 * Copyright (C) 2005, IBM Corp.
  6 *
  7 * Created by: Haren Myneni
  8 */
  9
 10#include <linux/kernel.h>
 11#include <linux/smp.h>
 12#include <linux/reboot.h>
 13#include <linux/kexec.h>
 14#include <linux/export.h>
 15#include <linux/crash_dump.h>
 16#include <linux/delay.h>
 17#include <linux/irq.h>
 18#include <linux/types.h>
 19
 20#include <asm/processor.h>
 21#include <asm/machdep.h>
 22#include <asm/kexec.h>
 23#include <asm/prom.h>
 24#include <asm/smp.h>
 25#include <asm/setjmp.h>
 26#include <asm/debug.h>
 27#include <asm/interrupt.h>
 28
 29/*
 30 * The primary CPU waits a while for all secondary CPUs to enter. This is to
 31 * avoid sending an IPI if the secondary CPUs are entering
 32 * crash_kexec_secondary on their own (eg via a system reset).
 33 *
 34 * The secondary timeout has to be longer than the primary. Both timeouts are
 35 * in milliseconds.
 36 */
 37#define PRIMARY_TIMEOUT		500
 38#define SECONDARY_TIMEOUT	1000
 39
 40#define IPI_TIMEOUT		10000
 41#define REAL_MODE_TIMEOUT	10000
 42
 43static int time_to_dump;
 44/*
 45 * crash_wake_offline should be set to 1 by platforms that intend to wake
 46 * up offline cpus prior to jumping to a kdump kernel. Currently powernv
 47 * sets it to 1, since we want to avoid things from happening when an
 48 * offline CPU wakes up due to something like an HMI (malfunction error),
 49 * which propagates to all threads.
 50 */
 51int crash_wake_offline;
 52
 53#define CRASH_HANDLER_MAX 3
 54/* List of shutdown handles */
 55static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX];
 56static DEFINE_SPINLOCK(crash_handlers_lock);
 57
 58static unsigned long crash_shutdown_buf[JMP_BUF_LEN];
 59static int crash_shutdown_cpu = -1;
 60
 61static int handle_fault(struct pt_regs *regs)
 62{
 63	if (crash_shutdown_cpu == smp_processor_id())
 64		longjmp(crash_shutdown_buf, 1);
 65	return 0;
 66}
 67
 68#ifdef CONFIG_SMP
 69
 70static atomic_t cpus_in_crash;
 71void crash_ipi_callback(struct pt_regs *regs)
 72{
 73	static cpumask_t cpus_state_saved = CPU_MASK_NONE;
 74
 75	int cpu = smp_processor_id();
 76
 77	hard_irq_disable();
 78	if (!cpumask_test_cpu(cpu, &cpus_state_saved)) {
 79		crash_save_cpu(regs, cpu);
 80		cpumask_set_cpu(cpu, &cpus_state_saved);
 81	}
 82
 83	atomic_inc(&cpus_in_crash);
 84	smp_mb__after_atomic();
 85
 86	/*
 87	 * Starting the kdump boot.
 88	 * This barrier is needed to make sure that all CPUs are stopped.
 89	 */
 90	while (!time_to_dump)
 91		cpu_relax();
 92
 93	if (ppc_md.kexec_cpu_down)
 94		ppc_md.kexec_cpu_down(1, 1);
 95
 96#ifdef CONFIG_PPC64
 97	kexec_smp_wait();
 98#else
 99	for (;;);	/* FIXME */
100#endif
101
102	/* NOTREACHED */
103}
104
105static void crash_kexec_prepare_cpus(int cpu)
106{
107	unsigned int msecs;
108	volatile unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
109	volatile int tries = 0;
110	int (*old_handler)(struct pt_regs *regs);
111
112	printk(KERN_EMERG "Sending IPI to other CPUs\n");
113
114	if (crash_wake_offline)
115		ncpus = num_present_cpus() - 1;
116
117	crash_send_ipi(crash_ipi_callback);
118	smp_wmb();
119
120again:
121	/*
122	 * FIXME: Until we will have the way to stop other CPUs reliably,
123	 * the crash CPU will send an IPI and wait for other CPUs to
124	 * respond.
125	 */
126	msecs = IPI_TIMEOUT;
127	while ((atomic_read(&cpus_in_crash) < ncpus) && (--msecs > 0))
128		mdelay(1);
129
130	/* Would it be better to replace the trap vector here? */
131
132	if (atomic_read(&cpus_in_crash) >= ncpus) {
133		printk(KERN_EMERG "IPI complete\n");
134		return;
135	}
136
137	printk(KERN_EMERG "ERROR: %d cpu(s) not responding\n",
138		ncpus - atomic_read(&cpus_in_crash));
139
140	/*
141	 * If we have a panic timeout set then we can't wait indefinitely
142	 * for someone to activate system reset. We also give up on the
143	 * second time through if system reset fail to work.
144	 */
145	if ((panic_timeout > 0) || (tries > 0))
146		return;
147
148	/*
149	 * A system reset will cause all CPUs to take an 0x100 exception.
150	 * The primary CPU returns here via setjmp, and the secondary
151	 * CPUs reexecute the crash_kexec_secondary path.
152	 */
153	old_handler = __debugger;
154	__debugger = handle_fault;
155	crash_shutdown_cpu = smp_processor_id();
156
157	if (setjmp(crash_shutdown_buf) == 0) {
158		printk(KERN_EMERG "Activate system reset (dumprestart) "
159				  "to stop other cpu(s)\n");
160
161		/*
162		 * A system reset will force all CPUs to execute the
163		 * crash code again. We need to reset cpus_in_crash so we
164		 * wait for everyone to do this.
165		 */
166		atomic_set(&cpus_in_crash, 0);
167		smp_mb();
168
169		while (atomic_read(&cpus_in_crash) < ncpus)
170			cpu_relax();
171	}
172
173	crash_shutdown_cpu = -1;
174	__debugger = old_handler;
175
176	tries++;
177	goto again;
178}
179
180/*
181 * This function will be called by secondary cpus.
182 */
183void crash_kexec_secondary(struct pt_regs *regs)
184{
185	unsigned long flags;
186	int msecs = SECONDARY_TIMEOUT;
187
188	local_irq_save(flags);
189
190	/* Wait for the primary crash CPU to signal its progress */
191	while (crashing_cpu < 0) {
192		if (--msecs < 0) {
193			/* No response, kdump image may not have been loaded */
194			local_irq_restore(flags);
195			return;
196		}
197
198		mdelay(1);
199	}
200
201	crash_ipi_callback(regs);
202}
203
204#else	/* ! CONFIG_SMP */
205
206static void crash_kexec_prepare_cpus(int cpu)
207{
208	/*
209	 * move the secondaries to us so that we can copy
210	 * the new kernel 0-0x100 safely
211	 *
212	 * do this if kexec in setup.c ?
213	 */
214#ifdef CONFIG_PPC64
215	smp_release_cpus();
216#else
217	/* FIXME */
218#endif
219}
220
221void crash_kexec_secondary(struct pt_regs *regs)
222{
223}
224#endif	/* CONFIG_SMP */
225
226/* wait for all the CPUs to hit real mode but timeout if they don't come in */
227#if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
228static void __maybe_unused crash_kexec_wait_realmode(int cpu)
229{
230	unsigned int msecs;
231	int i;
232
233	msecs = REAL_MODE_TIMEOUT;
234	for (i=0; i < nr_cpu_ids && msecs > 0; i++) {
235		if (i == cpu)
236			continue;
237
238		while (paca_ptrs[i]->kexec_state < KEXEC_STATE_REAL_MODE) {
239			barrier();
240			if (!cpu_possible(i) || !cpu_online(i) || (msecs <= 0))
241				break;
242			msecs--;
243			mdelay(1);
244		}
245	}
246	mb();
247}
248#else
249static inline void crash_kexec_wait_realmode(int cpu) {}
250#endif	/* CONFIG_SMP && CONFIG_PPC64 */
251
252/*
253 * Register a function to be called on shutdown.  Only use this if you
254 * can't reset your device in the second kernel.
255 */
256int crash_shutdown_register(crash_shutdown_t handler)
257{
258	unsigned int i, rc;
259
260	spin_lock(&crash_handlers_lock);
261	for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
262		if (!crash_shutdown_handles[i]) {
263			/* Insert handle at first empty entry */
264			crash_shutdown_handles[i] = handler;
265			rc = 0;
266			break;
267		}
268
269	if (i == CRASH_HANDLER_MAX) {
270		printk(KERN_ERR "Crash shutdown handles full, "
271		       "not registered.\n");
272		rc = 1;
273	}
274
275	spin_unlock(&crash_handlers_lock);
276	return rc;
277}
278EXPORT_SYMBOL(crash_shutdown_register);
279
280int crash_shutdown_unregister(crash_shutdown_t handler)
281{
282	unsigned int i, rc;
283
284	spin_lock(&crash_handlers_lock);
285	for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
286		if (crash_shutdown_handles[i] == handler)
287			break;
288
289	if (i == CRASH_HANDLER_MAX) {
290		printk(KERN_ERR "Crash shutdown handle not found\n");
291		rc = 1;
292	} else {
293		/* Shift handles down */
294		for (; i < (CRASH_HANDLER_MAX - 1); i++)
295			crash_shutdown_handles[i] =
296				crash_shutdown_handles[i+1];
297		/*
298		 * Reset last entry to NULL now that it has been shifted down,
299		 * this will allow new handles to be added here.
300		 */
301		crash_shutdown_handles[i] = NULL;
302		rc = 0;
303	}
304
305	spin_unlock(&crash_handlers_lock);
306	return rc;
307}
308EXPORT_SYMBOL(crash_shutdown_unregister);
309
310void default_machine_crash_shutdown(struct pt_regs *regs)
311{
312	unsigned int i;
313	int (*old_handler)(struct pt_regs *regs);
314
315	/* Avoid hardlocking with irresponsive CPU holding logbuf_lock */
316	printk_nmi_enter();
317
318	/*
319	 * This function is only called after the system
320	 * has panicked or is otherwise in a critical state.
321	 * The minimum amount of code to allow a kexec'd kernel
322	 * to run successfully needs to happen here.
323	 *
324	 * In practice this means stopping other cpus in
325	 * an SMP system.
326	 * The kernel is broken so disable interrupts.
327	 */
328	hard_irq_disable();
329
330	/*
331	 * Make a note of crashing cpu. Will be used in machine_kexec
332	 * such that another IPI will not be sent.
333	 */
334	crashing_cpu = smp_processor_id();
335
336	/*
337	 * If we came in via system reset, wait a while for the secondary
338	 * CPUs to enter.
339	 */
340	if (TRAP(regs) == INTERRUPT_SYSTEM_RESET)
341		mdelay(PRIMARY_TIMEOUT);
342
343	crash_kexec_prepare_cpus(crashing_cpu);
344
345	crash_save_cpu(regs, crashing_cpu);
346
347	time_to_dump = 1;
348
349	crash_kexec_wait_realmode(crashing_cpu);
350
351	machine_kexec_mask_interrupts();
352
353	/*
354	 * Call registered shutdown routines safely.  Swap out
355	 * __debugger_fault_handler, and replace on exit.
356	 */
357	old_handler = __debugger_fault_handler;
358	__debugger_fault_handler = handle_fault;
359	crash_shutdown_cpu = smp_processor_id();
360	for (i = 0; i < CRASH_HANDLER_MAX && crash_shutdown_handles[i]; i++) {
361		if (setjmp(crash_shutdown_buf) == 0) {
362			/*
363			 * Insert syncs and delay to ensure
364			 * instructions in the dangerous region don't
365			 * leak away from this protected region.
366			 */
367			asm volatile("sync; isync");
368			/* dangerous region */
369			crash_shutdown_handles[i]();
370			asm volatile("sync; isync");
371		}
372	}
373	crash_shutdown_cpu = -1;
374	__debugger_fault_handler = old_handler;
375
376	if (ppc_md.kexec_cpu_down)
377		ppc_md.kexec_cpu_down(1, 0);
378}