Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge tag 'x86-timers-2024-09-17' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull x86 timer updates from Thomas Gleixner:
- Use the topology information of number of packages for making the
decision about TSC trust instead of using the number of online nodes
which is not reflecting the real topology.
- Stop the PIT timer 0 when its not in use as to stop pointless
emulation in the VMM.
- Fix the PIT timer stop sequence for timer 0 so it truly stops both
real hardware and buggy VMM emulations.
* tag 'x86-timers-2024-09-17' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/tsc: Check for sockets instead of CPUs to make code match comment
clockevents/drivers/i8253: Fix stop sequence for timer 0
x86/i8253: Disable PIT timer 0 when not in use
x86/tsc: Use topology_max_packages() to get package number
+50
-37
-11
arch/x86/kernel/cpu/mshyperv.c
-11
arch/x86/kernel/cpu/mshyperv.c
···
16
16
#include <linux/interrupt.h>
17
17
#include <linux/irq.h>
18
18
#include <linux/kexec.h>
19
-
#include <linux/i8253.h>
20
19
#include <linux/random.h>
21
20
#include <asm/processor.h>
22
21
#include <asm/hypervisor.h>
···
535
536
*/
536
537
if (efi_enabled(EFI_BOOT))
537
538
x86_platform.get_nmi_reason = hv_get_nmi_reason;
538
-
539
-
/*
540
-
* Hyper-V VMs have a PIT emulation quirk such that zeroing the
541
-
* counter register during PIT shutdown restarts the PIT. So it
542
-
* continues to interrupt @18.2 HZ. Setting i8253_clear_counter
543
-
* to false tells pit_shutdown() not to zero the counter so that
544
-
* the PIT really is shutdown. Generation 2 VMs don't have a PIT,
545
-
* and setting this value has no effect.
546
-
*/
547
-
i8253_clear_counter_on_shutdown = false;
548
539
549
540
#if IS_ENABLED(CONFIG_HYPERV)
550
541
if ((hv_get_isolation_type() == HV_ISOLATION_TYPE_VBS) ||
+9
-2
arch/x86/kernel/i8253.c
+9
-2
arch/x86/kernel/i8253.c
···
8
8
#include <linux/timex.h>
9
9
#include <linux/i8253.h>
10
10
11
+
#include <asm/hypervisor.h>
11
12
#include <asm/apic.h>
12
13
#include <asm/hpet.h>
13
14
#include <asm/time.h>
···
40
39
41
40
bool __init pit_timer_init(void)
42
41
{
43
-
if (!use_pit())
42
+
if (!use_pit()) {
43
+
/*
44
+
* Don't just ignore the PIT. Ensure it's stopped, because
45
+
* VMMs otherwise steal CPU time just to pointlessly waggle
46
+
* the (masked) IRQ.
47
+
*/
48
+
clockevent_i8253_disable();
44
49
return false;
45
-
50
+
}
46
51
clockevent_i8253_init(true);
47
52
global_clock_event = &i8253_clockevent;
48
53
return true;
+4
-6
arch/x86/kernel/tsc.c
+4
-6
arch/x86/kernel/tsc.c
···
28
28
#include <asm/apic.h>
29
29
#include <asm/cpu_device_id.h>
30
30
#include <asm/i8259.h>
31
+
#include <asm/topology.h>
31
32
#include <asm/uv/uv.h>
32
33
33
34
unsigned int __read_mostly cpu_khz; /* TSC clocks / usec, not used here */
···
1254
1253
* - TSC which does not stop in C-States
1255
1254
* - the TSC_ADJUST register which allows to detect even minimal
1256
1255
* modifications
1257
-
* - not more than two sockets. As the number of sockets cannot be
1258
-
* evaluated at the early boot stage where this has to be
1259
-
* invoked, check the number of online memory nodes as a
1260
-
* fallback solution which is an reasonable estimate.
1256
+
* - not more than four packages
1261
1257
*/
1262
1258
if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
1263
1259
boot_cpu_has(X86_FEATURE_NONSTOP_TSC) &&
1264
1260
boot_cpu_has(X86_FEATURE_TSC_ADJUST) &&
1265
-
nr_online_nodes <= 4)
1261
+
topology_max_packages() <= 4)
1266
1262
tsc_disable_clocksource_watchdog();
1267
1263
}
1268
1264
···
1288
1290
*/
1289
1291
if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) {
1290
1292
/* assume multi socket systems are not synchronized: */
1291
-
if (num_possible_cpus() > 1)
1293
+
if (topology_max_packages() > 1)
1292
1294
return 1;
1293
1295
}
1294
1296
+36
-17
drivers/clocksource/i8253.c
+36
-17
drivers/clocksource/i8253.c
···
20
20
DEFINE_RAW_SPINLOCK(i8253_lock);
21
21
EXPORT_SYMBOL(i8253_lock);
22
22
23
-
/*
24
-
* Handle PIT quirk in pit_shutdown() where zeroing the counter register
25
-
* restarts the PIT, negating the shutdown. On platforms with the quirk,
26
-
* platform specific code can set this to false.
27
-
*/
28
-
bool i8253_clear_counter_on_shutdown __ro_after_init = true;
29
-
30
23
#ifdef CONFIG_CLKSRC_I8253
31
24
/*
32
25
* Since the PIT overflows every tick, its not very useful
···
101
108
#endif
102
109
103
110
#ifdef CONFIG_CLKEVT_I8253
111
+
void clockevent_i8253_disable(void)
112
+
{
113
+
raw_spin_lock(&i8253_lock);
114
+
115
+
/*
116
+
* Writing the MODE register should stop the counter, according to
117
+
* the datasheet. This appears to work on real hardware (well, on
118
+
* modern Intel and AMD boxes; I didn't dig the Pegasos out of the
119
+
* shed).
120
+
*
121
+
* However, some virtual implementations differ, and the MODE change
122
+
* doesn't have any effect until either the counter is written (KVM
123
+
* in-kernel PIT) or the next interrupt (QEMU). And in those cases,
124
+
* it may not stop the *count*, only the interrupts. Although in
125
+
* the virt case, that probably doesn't matter, as the value of the
126
+
* counter will only be calculated on demand if the guest reads it;
127
+
* it's the interrupts which cause steal time.
128
+
*
129
+
* Hyper-V apparently has a bug where even in mode 0, the IRQ keeps
130
+
* firing repeatedly if the counter is running. But it *does* do the
131
+
* right thing when the MODE register is written.
132
+
*
133
+
* So: write the MODE and then load the counter, which ensures that
134
+
* the IRQ is stopped on those buggy virt implementations. And then
135
+
* write the MODE again, which is the right way to stop it.
136
+
*/
137
+
outb_p(0x30, PIT_MODE);
138
+
outb_p(0, PIT_CH0);
139
+
outb_p(0, PIT_CH0);
140
+
141
+
outb_p(0x30, PIT_MODE);
142
+
143
+
raw_spin_unlock(&i8253_lock);
144
+
}
145
+
104
146
static int pit_shutdown(struct clock_event_device *evt)
105
147
{
106
148
if (!clockevent_state_oneshot(evt) && !clockevent_state_periodic(evt))
107
149
return 0;
108
150
109
-
raw_spin_lock(&i8253_lock);
110
-
111
-
outb_p(0x30, PIT_MODE);
112
-
113
-
if (i8253_clear_counter_on_shutdown) {
114
-
outb_p(0, PIT_CH0);
115
-
outb_p(0, PIT_CH0);
116
-
}
117
-
118
-
raw_spin_unlock(&i8253_lock);
151
+
clockevent_i8253_disable();
119
152
return 0;
120
153
}
121
154
+1
-1
include/linux/i8253.h
+1
-1
include/linux/i8253.h
···
21
21
#define PIT_LATCH ((PIT_TICK_RATE + HZ/2) / HZ)
22
22
23
23
extern raw_spinlock_t i8253_lock;
24
-
extern bool i8253_clear_counter_on_shutdown;
25
24
extern struct clock_event_device i8253_clockevent;
26
25
extern void clockevent_i8253_init(bool oneshot);
26
+
extern void clockevent_i8253_disable(void);
27
27
28
28
extern void setup_pit_timer(void);
29
29