blob: d798c0da451c80a2e249af32758ca86f926538a3 [file] [log] [blame]
Glauber Costa0941ecb2008-03-03 14:12:55 -03001/*
2 * Intel SMP support routines.
3 *
Alan Cox87c6fe22009-01-05 14:08:04 +00004 * (c) 1995 Alan Cox, Building #3 <alan@lxorguk.ukuu.org.uk>
Ingo Molnar8f47e162009-01-31 02:03:42 +01005 * (c) 1998-99, 2000, 2009 Ingo Molnar <mingo@redhat.com>
Glauber Costa0941ecb2008-03-03 14:12:55 -03006 * (c) 2002,2003 Andi Kleen, SuSE Labs.
7 *
8 * i386 and x86_64 integration by Glauber Costa <gcosta@redhat.com>
9 *
10 * This code is released under the GNU General Public License version 2 or
11 * later.
12 */
13
Glauber Costaf9e47a12008-03-03 14:12:52 -030014#include <linux/init.h>
15
16#include <linux/mm.h>
17#include <linux/delay.h>
18#include <linux/spinlock.h>
Paul Gortmaker69c60c82011-05-26 12:22:53 -040019#include <linux/export.h>
Glauber Costaf9e47a12008-03-03 14:12:52 -030020#include <linux/kernel_stat.h>
21#include <linux/mc146818rtc.h>
22#include <linux/cache.h>
23#include <linux/interrupt.h>
24#include <linux/cpu.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090025#include <linux/gfp.h>
Glauber Costaf9e47a12008-03-03 14:12:52 -030026
27#include <asm/mtrr.h>
28#include <asm/tlbflush.h>
29#include <asm/mmu_context.h>
30#include <asm/proto.h>
Ingo Molnar7b6aa332009-02-17 13:58:15 +010031#include <asm/apic.h>
Don Zickus7d007d22012-05-11 14:41:14 -040032#include <asm/nmi.h>
Ashok Raj8838eb62015-08-12 18:29:40 +020033#include <asm/mce.h>
Seiji Aguchicf910e82013-06-20 11:46:53 -040034#include <asm/trace/irq_vectors.h>
Hidehiro Kawai0ee59412016-10-11 13:54:23 -070035#include <asm/kexec.h>
Tiantian Fengfba4f472017-04-19 18:18:39 +020036#include <asm/virtext.h>
Hidehiro Kawai0ee59412016-10-11 13:54:23 -070037
Glauber Costa0941ecb2008-03-03 14:12:55 -030038/*
39 * Some notes on x86 processor bugs affecting SMP operation:
40 *
41 * Pentium, Pentium Pro, II, III (and all CPUs) have bugs.
42 * The Linux implications for SMP are handled as follows:
43 *
44 * Pentium III / [Xeon]
45 * None of the E1AP-E3AP errata are visible to the user.
46 *
47 * E1AP. see PII A1AP
48 * E2AP. see PII A2AP
49 * E3AP. see PII A3AP
50 *
51 * Pentium II / [Xeon]
52 * None of the A1AP-A3AP errata are visible to the user.
53 *
54 * A1AP. see PPro 1AP
55 * A2AP. see PPro 2AP
56 * A3AP. see PPro 7AP
57 *
58 * Pentium Pro
59 * None of 1AP-9AP errata are visible to the normal user,
60 * except occasional delivery of 'spurious interrupt' as trap #15.
61 * This is very rare and a non-problem.
62 *
63 * 1AP. Linux maps APIC as non-cacheable
64 * 2AP. worked around in hardware
65 * 3AP. fixed in C0 and above steppings microcode update.
66 * Linux does not use excessive STARTUP_IPIs.
67 * 4AP. worked around in hardware
68 * 5AP. symmetric IO mode (normal Linux operation) not affected.
69 * 'noapic' mode has vector 0xf filled out properly.
70 * 6AP. 'noapic' mode might be affected - fixed in later steppings
71 * 7AP. We do not assume writes to the LVT deassering IRQs
72 * 8AP. We do not enable low power mode (deep sleep) during MP bootup
73 * 9AP. We do not use mixed mode
74 *
75 * Pentium
76 * There is a marginal case where REP MOVS on 100MHz SMP
77 * machines with B stepping processors can fail. XXX should provide
78 * an L1cache=Writethrough or L1cache=off option.
79 *
80 * B stepping CPUs may hang. There are hardware work arounds
81 * for this. We warn about it in case your board doesn't have the work
82 * arounds. Basically that's so I can tell anyone with a B stepping
83 * CPU and SMP problems "tough".
84 *
85 * Specific items [From Pentium Processor Specification Update]
86 *
87 * 1AP. Linux doesn't use remote read
88 * 2AP. Linux doesn't trust APIC errors
89 * 3AP. We work around this
90 * 4AP. Linux never generated 3 interrupts of the same priority
91 * to cause a lost local interrupt.
92 * 5AP. Remote read is never used
93 * 6AP. not affected - worked around in hardware
94 * 7AP. not affected - worked around in hardware
95 * 8AP. worked around in hardware - we get explicit CS errors if not
96 * 9AP. only 'noapic' mode affected. Might generate spurious
97 * interrupts, we log only the first one and count the
98 * rest silently.
99 * 10AP. not affected - worked around in hardware
100 * 11AP. Linux reads the APIC between writes to avoid this, as per
101 * the documentation. Make sure you preserve this as it affects
102 * the C stepping chips too.
103 * 12AP. not affected - worked around in hardware
104 * 13AP. not affected - worked around in hardware
105 * 14AP. we always deassert INIT during bootup
106 * 15AP. not affected - worked around in hardware
107 * 16AP. not affected - worked around in hardware
108 * 17AP. not affected - worked around in hardware
109 * 18AP. not affected - worked around in hardware
110 * 19AP. not affected - worked around in BIOS
111 *
112 * If this sounds worrying believe me these bugs are either ___RARE___,
113 * or are signal timing bugs worked around in hardware and there's
114 * about nothing of note with C stepping upwards.
115 */
Glauber Costaf9e47a12008-03-03 14:12:52 -0300116
Don Zickus7d007d22012-05-11 14:41:14 -0400117static atomic_t stopping_cpu = ATOMIC_INIT(-1);
Don Zickus3aac27a2012-05-11 14:41:15 -0400118static bool smp_no_nmi_ipi = false;
Don Zickus7d007d22012-05-11 14:41:14 -0400119
Glauber Costaf9e47a12008-03-03 14:12:52 -0300120/*
121 * this function sends a 'reschedule' IPI to another CPU.
122 * it goes straight through and wastes no time serializing
123 * anything. Worst case is that we lose a reschedule ...
124 */
125static void native_smp_send_reschedule(int cpu)
126{
Gautham R Shenoyf6940102008-03-10 17:44:03 +0530127 if (unlikely(cpu_is_offline(cpu))) {
Prarit Bhargava21173d02017-04-18 08:25:05 -0400128 WARN(1, "sched: Unexpected reschedule of offline CPU#%d!\n", cpu);
Gautham R Shenoyf6940102008-03-10 17:44:03 +0530129 return;
130 }
Thomas Gleixner72613182015-11-04 22:57:09 +0000131 apic->send_IPI(cpu, RESCHEDULE_VECTOR);
Glauber Costaf9e47a12008-03-03 14:12:52 -0300132}
133
Jens Axboe3b16cf82008-06-26 11:21:54 +0200134void native_send_call_func_single_ipi(int cpu)
Glauber Costaf9e47a12008-03-03 14:12:52 -0300135{
Thomas Gleixner72613182015-11-04 22:57:09 +0000136 apic->send_IPI(cpu, CALL_FUNCTION_SINGLE_VECTOR);
Glauber Costaf9e47a12008-03-03 14:12:52 -0300137}
138
Mike Travisbcda0162008-12-16 17:33:59 -0800139void native_send_call_func_ipi(const struct cpumask *mask)
Glauber Costaf9e47a12008-03-03 14:12:52 -0300140{
Mike Travisc2d1cec2009-01-04 05:18:03 -0800141 cpumask_var_t allbutself;
Glauber Costaf9e47a12008-03-03 14:12:52 -0300142
Mike Travisc2d1cec2009-01-04 05:18:03 -0800143 if (!alloc_cpumask_var(&allbutself, GFP_ATOMIC)) {
Ingo Molnardac5f412009-01-28 15:42:24 +0100144 apic->send_IPI_mask(mask, CALL_FUNCTION_VECTOR);
Mike Travisc2d1cec2009-01-04 05:18:03 -0800145 return;
146 }
Glauber Costaf9e47a12008-03-03 14:12:52 -0300147
Mike Travisc2d1cec2009-01-04 05:18:03 -0800148 cpumask_copy(allbutself, cpu_online_mask);
149 cpumask_clear_cpu(smp_processor_id(), allbutself);
150
151 if (cpumask_equal(mask, allbutself) &&
152 cpumask_equal(cpu_online_mask, cpu_callout_mask))
Ingo Molnardac5f412009-01-28 15:42:24 +0100153 apic->send_IPI_allbutself(CALL_FUNCTION_VECTOR);
Glauber Costaf9e47a12008-03-03 14:12:52 -0300154 else
Ingo Molnardac5f412009-01-28 15:42:24 +0100155 apic->send_IPI_mask(mask, CALL_FUNCTION_VECTOR);
Mike Travisc2d1cec2009-01-04 05:18:03 -0800156
157 free_cpumask_var(allbutself);
Glauber Costaf9e47a12008-03-03 14:12:52 -0300158}
159
Don Zickus7d007d22012-05-11 14:41:14 -0400160static int smp_stop_nmi_callback(unsigned int val, struct pt_regs *regs)
161{
162 /* We are registered on stopping cpu too, avoid spurious NMI */
163 if (raw_smp_processor_id() == atomic_read(&stopping_cpu))
164 return NMI_HANDLED;
165
Tiantian Fengfba4f472017-04-19 18:18:39 +0200166 cpu_emergency_vmxoff();
Don Zickus7d007d22012-05-11 14:41:14 -0400167 stop_this_cpu(NULL);
168
169 return NMI_HANDLED;
170}
171
Glauber Costaf9e47a12008-03-03 14:12:52 -0300172/*
173 * this function calls the 'stop' function on all other CPUs in the system.
174 */
175
Andi Kleen2605fc22014-05-02 00:44:37 +0200176asmlinkage __visible void smp_reboot_interrupt(void)
Andi Kleen4ef702c2009-05-27 21:56:52 +0200177{
Thomas Gleixner6dc17872015-05-15 15:50:45 +0200178 ipi_entering_ack_irq();
Tiantian Fengfba4f472017-04-19 18:18:39 +0200179 cpu_emergency_vmxoff();
Andi Kleen4ef702c2009-05-27 21:56:52 +0200180 stop_this_cpu(NULL);
181 irq_exit();
182}
183
Don Zickus5d2b86d2012-05-11 14:41:13 -0400184static void native_stop_other_cpus(int wait)
Glauber Costaf9e47a12008-03-03 14:12:52 -0300185{
Glauber Costaf9e47a12008-03-03 14:12:52 -0300186 unsigned long flags;
Alok Kataria76fac072010-10-11 14:37:08 -0700187 unsigned long timeout;
Glauber Costaf9e47a12008-03-03 14:12:52 -0300188
189 if (reboot_force)
190 return;
191
Andi Kleen4ef702c2009-05-27 21:56:52 +0200192 /*
193 * Use an own vector here because smp_call_function
194 * does lots of things not suitable in a panic situation.
Don Zickus7d007d22012-05-11 14:41:14 -0400195 */
196
197 /*
198 * We start by using the REBOOT_VECTOR irq.
199 * The irq is treated as a sync point to allow critical
200 * regions of code on other cpus to release their spin locks
201 * and re-enable irqs. Jumping straight to an NMI might
202 * accidentally cause deadlocks with further shutdown/panic
203 * code. By syncing, we give the cpus up to one second to
204 * finish their work before we force them off with the NMI.
Andi Kleen4ef702c2009-05-27 21:56:52 +0200205 */
206 if (num_online_cpus() > 1) {
Don Zickus7d007d22012-05-11 14:41:14 -0400207 /* did someone beat us here? */
208 if (atomic_cmpxchg(&stopping_cpu, -1, safe_smp_processor_id()) != -1)
209 return;
210
211 /* sync above data before sending IRQ */
212 wmb();
213
Andi Kleen4ef702c2009-05-27 21:56:52 +0200214 apic->send_IPI_allbutself(REBOOT_VECTOR);
215
Alok Kataria76fac072010-10-11 14:37:08 -0700216 /*
217 * Don't wait longer than a second if the caller
218 * didn't ask us to wait.
219 */
220 timeout = USEC_PER_SEC;
221 while (num_online_cpus() > 1 && (wait || timeout--))
Andi Kleen4ef702c2009-05-27 21:56:52 +0200222 udelay(1);
223 }
Don Zickus7d007d22012-05-11 14:41:14 -0400224
225 /* if the REBOOT_VECTOR didn't work, try with the NMI */
Don Zickus3aac27a2012-05-11 14:41:15 -0400226 if ((num_online_cpus() > 1) && (!smp_no_nmi_ipi)) {
Don Zickus7d007d22012-05-11 14:41:14 -0400227 if (register_nmi_handler(NMI_LOCAL, smp_stop_nmi_callback,
228 NMI_FLAG_FIRST, "smp_stop"))
229 /* Note: we ignore failures here */
230 /* Hope the REBOOT_IRQ is good enough */
231 goto finish;
Andi Kleen4ef702c2009-05-27 21:56:52 +0200232
Don Zickus7d007d22012-05-11 14:41:14 -0400233 /* sync above data before sending IRQ */
234 wmb();
235
236 pr_emerg("Shutting down cpus with NMI\n");
237
238 apic->send_IPI_allbutself(NMI_VECTOR);
239
240 /*
241 * Don't wait longer than a 10 ms if the caller
242 * didn't ask us to wait.
243 */
244 timeout = USEC_PER_MSEC * 10;
245 while (num_online_cpus() > 1 && (wait || timeout--))
246 udelay(1);
247 }
248
249finish:
Glauber Costaf9e47a12008-03-03 14:12:52 -0300250 local_irq_save(flags);
Glauber Costaf9e47a12008-03-03 14:12:52 -0300251 disable_local_APIC();
Ashok Raj8838eb62015-08-12 18:29:40 +0200252 mcheck_cpu_clear(this_cpu_ptr(&cpu_info));
Glauber Costaf9e47a12008-03-03 14:12:52 -0300253 local_irq_restore(flags);
254}
255
256/*
Peter Zijlstra184748c2011-04-05 17:23:39 +0200257 * Reschedule call back.
Glauber Costaf9e47a12008-03-03 14:12:52 -0300258 */
Seiji Aguchieddc0e92013-06-20 11:45:17 -0400259static inline void __smp_reschedule_interrupt(void)
260{
261 inc_irq_stat(irq_resched_count);
262 scheduler_ipi();
263}
264
Daniel Bristot de Oliveirac4158ff2017-01-04 12:20:33 +0100265__visible void __irq_entry smp_reschedule_interrupt(struct pt_regs *regs)
Glauber Costaf9e47a12008-03-03 14:12:52 -0300266{
267 ack_APIC_irq();
Seiji Aguchieddc0e92013-06-20 11:45:17 -0400268 __smp_reschedule_interrupt();
Marcelo Tosatti32f88402009-05-07 17:55:12 -0300269 /*
270 * KVM uses this interrupt to force a cpu out of guest mode
271 */
Glauber Costaf9e47a12008-03-03 14:12:52 -0300272}
273
Daniel Bristot de Oliveirac4158ff2017-01-04 12:20:33 +0100274__visible void __irq_entry smp_trace_reschedule_interrupt(struct pt_regs *regs)
Seiji Aguchi4787c362013-06-28 14:02:11 -0400275{
276 /*
277 * Need to call irq_enter() before calling the trace point.
278 * __smp_reschedule_interrupt() calls irq_enter/exit() too (in
279 * scheduler_ipi(). This is OK, since those functions are allowed
280 * to nest.
281 */
Thomas Gleixner6dc17872015-05-15 15:50:45 +0200282 ipi_entering_ack_irq();
Seiji Aguchi4787c362013-06-28 14:02:11 -0400283 trace_reschedule_entry(RESCHEDULE_VECTOR);
284 __smp_reschedule_interrupt();
285 trace_reschedule_exit(RESCHEDULE_VECTOR);
286 exiting_irq();
287 /*
288 * KVM uses this interrupt to force a cpu out of guest mode
289 */
290}
291
Seiji Aguchieddc0e92013-06-20 11:45:17 -0400292static inline void __smp_call_function_interrupt(void)
293{
Jens Axboe3b16cf82008-06-26 11:21:54 +0200294 generic_smp_call_function_interrupt();
Hiroshi Shimamoto915b0d02008-12-08 19:19:26 -0800295 inc_irq_stat(irq_call_count);
Seiji Aguchieddc0e92013-06-20 11:45:17 -0400296}
297
Daniel Bristot de Oliveirac4158ff2017-01-04 12:20:33 +0100298__visible void __irq_entry smp_call_function_interrupt(struct pt_regs *regs)
Seiji Aguchieddc0e92013-06-20 11:45:17 -0400299{
Thomas Gleixner6dc17872015-05-15 15:50:45 +0200300 ipi_entering_ack_irq();
Seiji Aguchieddc0e92013-06-20 11:45:17 -0400301 __smp_call_function_interrupt();
302 exiting_irq();
303}
304
Daniel Bristot de Oliveirac4158ff2017-01-04 12:20:33 +0100305__visible void __irq_entry
306smp_trace_call_function_interrupt(struct pt_regs *regs)
Seiji Aguchicf910e82013-06-20 11:46:53 -0400307{
Thomas Gleixner6dc17872015-05-15 15:50:45 +0200308 ipi_entering_ack_irq();
Seiji Aguchicf910e82013-06-20 11:46:53 -0400309 trace_call_function_entry(CALL_FUNCTION_VECTOR);
310 __smp_call_function_interrupt();
311 trace_call_function_exit(CALL_FUNCTION_VECTOR);
312 exiting_irq();
313}
314
Seiji Aguchieddc0e92013-06-20 11:45:17 -0400315static inline void __smp_call_function_single_interrupt(void)
316{
317 generic_smp_call_function_single_interrupt();
318 inc_irq_stat(irq_call_count);
Jens Axboe3b16cf82008-06-26 11:21:54 +0200319}
Glauber Costaf9e47a12008-03-03 14:12:52 -0300320
Daniel Bristot de Oliveirac4158ff2017-01-04 12:20:33 +0100321__visible void __irq_entry
322smp_call_function_single_interrupt(struct pt_regs *regs)
Jens Axboe3b16cf82008-06-26 11:21:54 +0200323{
Thomas Gleixner6dc17872015-05-15 15:50:45 +0200324 ipi_entering_ack_irq();
Seiji Aguchieddc0e92013-06-20 11:45:17 -0400325 __smp_call_function_single_interrupt();
326 exiting_irq();
Glauber Costaf9e47a12008-03-03 14:12:52 -0300327}
328
Daniel Bristot de Oliveirac4158ff2017-01-04 12:20:33 +0100329__visible void __irq_entry
330smp_trace_call_function_single_interrupt(struct pt_regs *regs)
Seiji Aguchicf910e82013-06-20 11:46:53 -0400331{
Thomas Gleixner6dc17872015-05-15 15:50:45 +0200332 ipi_entering_ack_irq();
Seiji Aguchicf910e82013-06-20 11:46:53 -0400333 trace_call_function_single_entry(CALL_FUNCTION_SINGLE_VECTOR);
334 __smp_call_function_single_interrupt();
335 trace_call_function_single_exit(CALL_FUNCTION_SINGLE_VECTOR);
336 exiting_irq();
337}
338
Don Zickusbda62632011-10-13 15:14:27 -0400339static int __init nonmi_ipi_setup(char *str)
340{
Don Zickus3aac27a2012-05-11 14:41:15 -0400341 smp_no_nmi_ipi = true;
342 return 1;
Don Zickusbda62632011-10-13 15:14:27 -0400343}
344
345__setup("nonmi_ipi", nonmi_ipi_setup);
346
Glauber Costaf9e47a12008-03-03 14:12:52 -0300347struct smp_ops smp_ops = {
Cyrill Gorcunovb9b34f22009-04-12 20:47:42 +0400348 .smp_prepare_boot_cpu = native_smp_prepare_boot_cpu,
349 .smp_prepare_cpus = native_smp_prepare_cpus,
350 .smp_cpus_done = native_smp_cpus_done,
Glauber Costaf9e47a12008-03-03 14:12:52 -0300351
Don Zickus5d2b86d2012-05-11 14:41:13 -0400352 .stop_other_cpus = native_stop_other_cpus,
Hidehiro Kawai0ee59412016-10-11 13:54:23 -0700353#if defined(CONFIG_KEXEC_CORE)
354 .crash_stop_other_cpus = kdump_nmi_shootdown_cpus,
355#endif
Cyrill Gorcunovb9b34f22009-04-12 20:47:42 +0400356 .smp_send_reschedule = native_smp_send_reschedule,
Jens Axboe3b16cf82008-06-26 11:21:54 +0200357
Cyrill Gorcunovb9b34f22009-04-12 20:47:42 +0400358 .cpu_up = native_cpu_up,
359 .cpu_die = native_cpu_die,
360 .cpu_disable = native_cpu_disable,
361 .play_dead = native_play_dead,
Alex Nixon93be71b2008-08-22 11:52:11 +0100362
Cyrill Gorcunovb9b34f22009-04-12 20:47:42 +0400363 .send_call_func_ipi = native_send_call_func_ipi,
Jens Axboe3b16cf82008-06-26 11:21:54 +0200364 .send_call_func_single_ipi = native_send_call_func_single_ipi,
Glauber Costaf9e47a12008-03-03 14:12:52 -0300365};
366EXPORT_SYMBOL_GPL(smp_ops);