xref: /openbmc/linux/arch/x86/kernel/smp.c (revision 050bb587)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *	Intel SMP support routines.
4  *
5  *	(c) 1995 Alan Cox, Building #3 <alan@lxorguk.ukuu.org.uk>
6  *	(c) 1998-99, 2000, 2009 Ingo Molnar <mingo@redhat.com>
7  *      (c) 2002,2003 Andi Kleen, SuSE Labs.
8  *
9  *	i386 and x86_64 integration by Glauber Costa <gcosta@redhat.com>
10  */
11 
12 #include <linux/init.h>
13 
14 #include <linux/mm.h>
15 #include <linux/delay.h>
16 #include <linux/spinlock.h>
17 #include <linux/export.h>
18 #include <linux/kernel_stat.h>
19 #include <linux/mc146818rtc.h>
20 #include <linux/cache.h>
21 #include <linux/interrupt.h>
22 #include <linux/cpu.h>
23 #include <linux/gfp.h>
24 #include <linux/kexec.h>
25 
26 #include <asm/mtrr.h>
27 #include <asm/tlbflush.h>
28 #include <asm/mmu_context.h>
29 #include <asm/proto.h>
30 #include <asm/apic.h>
31 #include <asm/cpu.h>
32 #include <asm/idtentry.h>
33 #include <asm/nmi.h>
34 #include <asm/mce.h>
35 #include <asm/trace/irq_vectors.h>
36 #include <asm/kexec.h>
37 #include <asm/reboot.h>
38 
39 /*
40  *	Some notes on x86 processor bugs affecting SMP operation:
41  *
42  *	Pentium, Pentium Pro, II, III (and all CPUs) have bugs.
43  *	The Linux implications for SMP are handled as follows:
44  *
45  *	Pentium III / [Xeon]
46  *		None of the E1AP-E3AP errata are visible to the user.
47  *
48  *	E1AP.	see PII A1AP
49  *	E2AP.	see PII A2AP
50  *	E3AP.	see PII A3AP
51  *
52  *	Pentium II / [Xeon]
53  *		None of the A1AP-A3AP errata are visible to the user.
54  *
55  *	A1AP.	see PPro 1AP
56  *	A2AP.	see PPro 2AP
57  *	A3AP.	see PPro 7AP
58  *
59  *	Pentium Pro
60  *		None of 1AP-9AP errata are visible to the normal user,
61  *	except occasional delivery of 'spurious interrupt' as trap #15.
62  *	This is very rare and a non-problem.
63  *
64  *	1AP.	Linux maps APIC as non-cacheable
65  *	2AP.	worked around in hardware
66  *	3AP.	fixed in C0 and above steppings microcode update.
67  *		Linux does not use excessive STARTUP_IPIs.
68  *	4AP.	worked around in hardware
69  *	5AP.	symmetric IO mode (normal Linux operation) not affected.
70  *		'noapic' mode has vector 0xf filled out properly.
71  *	6AP.	'noapic' mode might be affected - fixed in later steppings
72  *	7AP.	We do not assume writes to the LVT deasserting IRQs
73  *	8AP.	We do not enable low power mode (deep sleep) during MP bootup
74  *	9AP.	We do not use mixed mode
75  *
76  *	Pentium
77  *		There is a marginal case where REP MOVS on 100MHz SMP
78  *	machines with B stepping processors can fail. XXX should provide
79  *	an L1cache=Writethrough or L1cache=off option.
80  *
81  *		B stepping CPUs may hang. There are hardware work arounds
82  *	for this. We warn about it in case your board doesn't have the work
83  *	arounds. Basically that's so I can tell anyone with a B stepping
84  *	CPU and SMP problems "tough".
85  *
86  *	Specific items [From Pentium Processor Specification Update]
87  *
88  *	1AP.	Linux doesn't use remote read
89  *	2AP.	Linux doesn't trust APIC errors
90  *	3AP.	We work around this
91  *	4AP.	Linux never generated 3 interrupts of the same priority
92  *		to cause a lost local interrupt.
93  *	5AP.	Remote read is never used
94  *	6AP.	not affected - worked around in hardware
95  *	7AP.	not affected - worked around in hardware
96  *	8AP.	worked around in hardware - we get explicit CS errors if not
97  *	9AP.	only 'noapic' mode affected. Might generate spurious
98  *		interrupts, we log only the first one and count the
99  *		rest silently.
100  *	10AP.	not affected - worked around in hardware
101  *	11AP.	Linux reads the APIC between writes to avoid this, as per
102  *		the documentation. Make sure you preserve this as it affects
103  *		the C stepping chips too.
104  *	12AP.	not affected - worked around in hardware
105  *	13AP.	not affected - worked around in hardware
106  *	14AP.	we always deassert INIT during bootup
107  *	15AP.	not affected - worked around in hardware
108  *	16AP.	not affected - worked around in hardware
109  *	17AP.	not affected - worked around in hardware
110  *	18AP.	not affected - worked around in hardware
111  *	19AP.	not affected - worked around in BIOS
112  *
113  *	If this sounds worrying believe me these bugs are either ___RARE___,
114  *	or are signal timing bugs worked around in hardware and there's
115  *	about nothing of note with C stepping upwards.
116  */
117 
118 static atomic_t stopping_cpu = ATOMIC_INIT(-1);
119 static bool smp_no_nmi_ipi = false;
120 
121 static int smp_stop_nmi_callback(unsigned int val, struct pt_regs *regs)
122 {
123 	/* We are registered on stopping cpu too, avoid spurious NMI */
124 	if (raw_smp_processor_id() == atomic_read(&stopping_cpu))
125 		return NMI_HANDLED;
126 
127 	cpu_emergency_disable_virtualization();
128 	stop_this_cpu(NULL);
129 
130 	return NMI_HANDLED;
131 }
132 
133 /*
134  * Disable virtualization, APIC etc. and park the CPU in a HLT loop
135  */
136 DEFINE_IDTENTRY_SYSVEC(sysvec_reboot)
137 {
138 	ack_APIC_irq();
139 	cpu_emergency_disable_virtualization();
140 	stop_this_cpu(NULL);
141 }
142 
143 static int register_stop_handler(void)
144 {
145 	return register_nmi_handler(NMI_LOCAL, smp_stop_nmi_callback,
146 				    NMI_FLAG_FIRST, "smp_stop");
147 }
148 
149 static void native_stop_other_cpus(int wait)
150 {
151 	unsigned int cpu = smp_processor_id();
152 	unsigned long flags, timeout;
153 
154 	if (reboot_force)
155 		return;
156 
157 	/* Only proceed if this is the first CPU to reach this code */
158 	if (atomic_cmpxchg(&stopping_cpu, -1, cpu) != -1)
159 		return;
160 
161 	/* For kexec, ensure that offline CPUs are out of MWAIT and in HLT */
162 	if (kexec_in_progress)
163 		smp_kick_mwait_play_dead();
164 
165 	/*
166 	 * 1) Send an IPI on the reboot vector to all other CPUs.
167 	 *
168 	 *    The other CPUs should react on it after leaving critical
169 	 *    sections and re-enabling interrupts. They might still hold
170 	 *    locks, but there is nothing which can be done about that.
171 	 *
172 	 * 2) Wait for all other CPUs to report that they reached the
173 	 *    HLT loop in stop_this_cpu()
174 	 *
175 	 * 3) If the system uses INIT/STARTUP for CPU bringup, then
176 	 *    send all present CPUs an INIT vector, which brings them
177 	 *    completely out of the way.
178 	 *
179 	 * 4) If #3 is not possible and #2 timed out send an NMI to the
180 	 *    CPUs which did not yet report
181 	 *
182 	 * 5) Wait for all other CPUs to report that they reached the
183 	 *    HLT loop in stop_this_cpu()
184 	 *
185 	 * #4 can obviously race against a CPU reaching the HLT loop late.
186 	 * That CPU will have reported already and the "have all CPUs
187 	 * reached HLT" condition will be true despite the fact that the
188 	 * other CPU is still handling the NMI. Again, there is no
189 	 * protection against that as "disabled" APICs still respond to
190 	 * NMIs.
191 	 */
192 	cpumask_copy(&cpus_stop_mask, cpu_online_mask);
193 	cpumask_clear_cpu(cpu, &cpus_stop_mask);
194 
195 	if (!cpumask_empty(&cpus_stop_mask)) {
196 		apic_send_IPI_allbutself(REBOOT_VECTOR);
197 
198 		/*
199 		 * Don't wait longer than a second for IPI completion. The
200 		 * wait request is not checked here because that would
201 		 * prevent an NMI/INIT shutdown in case that not all
202 		 * CPUs reach shutdown state.
203 		 */
204 		timeout = USEC_PER_SEC;
205 		while (!cpumask_empty(&cpus_stop_mask) && timeout--)
206 			udelay(1);
207 	}
208 
209 	/*
210 	 * Park all other CPUs in INIT including "offline" CPUs, if
211 	 * possible. That's a safe place where they can't resume execution
212 	 * of HLT and then execute the HLT loop from overwritten text or
213 	 * page tables.
214 	 *
215 	 * The only downside is a broadcast MCE, but up to the point where
216 	 * the kexec() kernel brought all APs online again an MCE will just
217 	 * make HLT resume and handle the MCE. The machine crashes and burns
218 	 * due to overwritten text, page tables and data. So there is a
219 	 * choice between fire and frying pan. The result is pretty much
220 	 * the same. Chose frying pan until x86 provides a sane mechanism
221 	 * to park a CPU.
222 	 */
223 	if (smp_park_other_cpus_in_init())
224 		goto done;
225 
226 	/*
227 	 * If park with INIT was not possible and the REBOOT_VECTOR didn't
228 	 * take all secondary CPUs offline, try with the NMI.
229 	 */
230 	if (!cpumask_empty(&cpus_stop_mask)) {
231 		/*
232 		 * If NMI IPI is enabled, try to register the stop handler
233 		 * and send the IPI. In any case try to wait for the other
234 		 * CPUs to stop.
235 		 */
236 		if (!smp_no_nmi_ipi && !register_stop_handler()) {
237 			pr_emerg("Shutting down cpus with NMI\n");
238 
239 			for_each_cpu(cpu, &cpus_stop_mask)
240 				apic->send_IPI(cpu, NMI_VECTOR);
241 		}
242 		/*
243 		 * Don't wait longer than 10 ms if the caller didn't
244 		 * request it. If wait is true, the machine hangs here if
245 		 * one or more CPUs do not reach shutdown state.
246 		 */
247 		timeout = USEC_PER_MSEC * 10;
248 		while (!cpumask_empty(&cpus_stop_mask) && (wait || timeout--))
249 			udelay(1);
250 	}
251 
252 done:
253 	local_irq_save(flags);
254 	disable_local_APIC();
255 	mcheck_cpu_clear(this_cpu_ptr(&cpu_info));
256 	local_irq_restore(flags);
257 
258 	/*
259 	 * Ensure that the cpus_stop_mask cache lines are invalidated on
260 	 * the other CPUs. See comment vs. SME in stop_this_cpu().
261 	 */
262 	cpumask_clear(&cpus_stop_mask);
263 }
264 
265 /*
266  * Reschedule call back. KVM uses this interrupt to force a cpu out of
267  * guest mode.
268  */
269 DEFINE_IDTENTRY_SYSVEC_SIMPLE(sysvec_reschedule_ipi)
270 {
271 	ack_APIC_irq();
272 	trace_reschedule_entry(RESCHEDULE_VECTOR);
273 	inc_irq_stat(irq_resched_count);
274 	scheduler_ipi();
275 	trace_reschedule_exit(RESCHEDULE_VECTOR);
276 }
277 
278 DEFINE_IDTENTRY_SYSVEC(sysvec_call_function)
279 {
280 	ack_APIC_irq();
281 	trace_call_function_entry(CALL_FUNCTION_VECTOR);
282 	inc_irq_stat(irq_call_count);
283 	generic_smp_call_function_interrupt();
284 	trace_call_function_exit(CALL_FUNCTION_VECTOR);
285 }
286 
287 DEFINE_IDTENTRY_SYSVEC(sysvec_call_function_single)
288 {
289 	ack_APIC_irq();
290 	trace_call_function_single_entry(CALL_FUNCTION_SINGLE_VECTOR);
291 	inc_irq_stat(irq_call_count);
292 	generic_smp_call_function_single_interrupt();
293 	trace_call_function_single_exit(CALL_FUNCTION_SINGLE_VECTOR);
294 }
295 
296 static int __init nonmi_ipi_setup(char *str)
297 {
298 	smp_no_nmi_ipi = true;
299 	return 1;
300 }
301 
302 __setup("nonmi_ipi", nonmi_ipi_setup);
303 
304 struct smp_ops smp_ops = {
305 	.smp_prepare_boot_cpu	= native_smp_prepare_boot_cpu,
306 	.smp_prepare_cpus	= native_smp_prepare_cpus,
307 	.smp_cpus_done		= native_smp_cpus_done,
308 
309 	.stop_other_cpus	= native_stop_other_cpus,
310 #if defined(CONFIG_KEXEC_CORE)
311 	.crash_stop_other_cpus	= kdump_nmi_shootdown_cpus,
312 #endif
313 	.smp_send_reschedule	= native_smp_send_reschedule,
314 
315 	.kick_ap_alive		= native_kick_ap,
316 	.cpu_disable		= native_cpu_disable,
317 	.play_dead		= native_play_dead,
318 
319 	.send_call_func_ipi	= native_send_call_func_ipi,
320 	.send_call_func_single_ipi = native_send_call_func_single_ipi,
321 };
322 EXPORT_SYMBOL_GPL(smp_ops);
323