1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _ASM_X86_MWAIT_H
3#define _ASM_X86_MWAIT_H
4
5#include <linux/sched.h>
6#include <linux/sched/idle.h>
7
8#include <asm/cpufeature.h>
9#include <asm/nospec-branch.h>
10
11#define MWAIT_SUBSTATE_MASK		0xf
12#define MWAIT_CSTATE_MASK		0xf
13#define MWAIT_SUBSTATE_SIZE		4
14#define MWAIT_HINT2CSTATE(hint)		(((hint) >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK)
15#define MWAIT_HINT2SUBSTATE(hint)	((hint) & MWAIT_CSTATE_MASK)
16#define MWAIT_C1_SUBSTATE_MASK  0xf0
17
18#define CPUID_MWAIT_LEAF		5
19#define CPUID5_ECX_EXTENSIONS_SUPPORTED 0x1
20#define CPUID5_ECX_INTERRUPT_BREAK	0x2
21
22#define MWAIT_ECX_INTERRUPT_BREAK	0x1
23#define MWAITX_ECX_TIMER_ENABLE		BIT(1)
24#define MWAITX_MAX_WAIT_CYCLES		UINT_MAX
25#define MWAITX_DISABLE_CSTATES		0xf0
26#define TPAUSE_C01_STATE		1
27#define TPAUSE_C02_STATE		0
28
29static __always_inline void __monitor(const void *eax, unsigned long ecx,
30			     unsigned long edx)
31{
32	/* "monitor %eax, %ecx, %edx;" */
33	asm volatile(".byte 0x0f, 0x01, 0xc8;"
34		     :: "a" (eax), "c" (ecx), "d"(edx));
35}
36
37static __always_inline void __monitorx(const void *eax, unsigned long ecx,
38			      unsigned long edx)
39{
40	/* "monitorx %eax, %ecx, %edx;" */
41	asm volatile(".byte 0x0f, 0x01, 0xfa;"
42		     :: "a" (eax), "c" (ecx), "d"(edx));
43}
44
45static __always_inline void __mwait(unsigned long eax, unsigned long ecx)
46{
47	mds_idle_clear_cpu_buffers();
48
49	/* "mwait %eax, %ecx;" */
50	asm volatile(".byte 0x0f, 0x01, 0xc9;"
51		     :: "a" (eax), "c" (ecx));
52}
53
54/*
55 * MWAITX allows for a timer expiration to get the core out a wait state in
56 * addition to the default MWAIT exit condition of a store appearing at a
57 * monitored virtual address.
58 *
59 * Registers:
60 *
61 * MWAITX ECX[1]: enable timer if set
62 * MWAITX EBX[31:0]: max wait time expressed in SW P0 clocks. The software P0
63 * frequency is the same as the TSC frequency.
64 *
65 * Below is a comparison between MWAIT and MWAITX on AMD processors:
66 *
67 *                 MWAIT                           MWAITX
68 * opcode          0f 01 c9           |            0f 01 fb
69 * ECX[0]                  value of RFLAGS.IF seen by instruction
70 * ECX[1]          unused/#GP if set  |            enable timer if set
71 * ECX[31:2]                     unused/#GP if set
72 * EAX                           unused (reserve for hint)
73 * EBX[31:0]       unused             |            max wait time (P0 clocks)
74 *
75 *                 MONITOR                         MONITORX
76 * opcode          0f 01 c8           |            0f 01 fa
77 * EAX                     (logical) address to monitor
78 * ECX                     #GP if not zero
79 */
80static __always_inline void __mwaitx(unsigned long eax, unsigned long ebx,
81				     unsigned long ecx)
82{
83	/* No MDS buffer clear as this is AMD/HYGON only */
84
85	/* "mwaitx %eax, %ebx, %ecx;" */
86	asm volatile(".byte 0x0f, 0x01, 0xfb;"
87		     :: "a" (eax), "b" (ebx), "c" (ecx));
88}
89
90/*
91 * Re-enable interrupts right upon calling mwait in such a way that
92 * no interrupt can fire _before_ the execution of mwait, ie: no
93 * instruction must be placed between "sti" and "mwait".
94 *
95 * This is necessary because if an interrupt queues a timer before
96 * executing mwait, it would otherwise go unnoticed and the next tick
97 * would not be reprogrammed accordingly before mwait ever wakes up.
98 */
99static __always_inline void __sti_mwait(unsigned long eax, unsigned long ecx)
100{
101	mds_idle_clear_cpu_buffers();
102	/* "mwait %eax, %ecx;" */
103	asm volatile("sti; .byte 0x0f, 0x01, 0xc9;"
104		     :: "a" (eax), "c" (ecx));
105}
106
107/*
108 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
109 * which can obviate IPI to trigger checking of need_resched.
110 * We execute MONITOR against need_resched and enter optimized wait state
111 * through MWAIT. Whenever someone changes need_resched, we would be woken
112 * up from MWAIT (without an IPI).
113 *
114 * New with Core Duo processors, MWAIT can take some hints based on CPU
115 * capability.
116 */
117static __always_inline void mwait_idle_with_hints(unsigned long eax, unsigned long ecx)
118{
119	if (static_cpu_has_bug(X86_BUG_MONITOR) || !current_set_polling_and_test()) {
120		if (static_cpu_has_bug(X86_BUG_CLFLUSH_MONITOR)) {
121			mb();
122			clflush((void *)&current_thread_info()->flags);
123			mb();
124		}
125
126		__monitor((void *)&current_thread_info()->flags, 0, 0);
127
128		if (!need_resched()) {
129			if (ecx & 1) {
130				__mwait(eax, ecx);
131			} else {
132				__sti_mwait(eax, ecx);
133				raw_local_irq_disable();
134			}
135		}
136	}
137	current_clr_polling();
138}
139
140/*
141 * Caller can specify whether to enter C0.1 (low latency, less
142 * power saving) or C0.2 state (saves more power, but longer wakeup
143 * latency). This may be overridden by the IA32_UMWAIT_CONTROL MSR
144 * which can force requests for C0.2 to be downgraded to C0.1.
145 */
146static inline void __tpause(u32 ecx, u32 edx, u32 eax)
147{
148	/* "tpause %ecx, %edx, %eax;" */
149	#ifdef CONFIG_AS_TPAUSE
150	asm volatile("tpause %%ecx\n"
151		     :
152		     : "c"(ecx), "d"(edx), "a"(eax));
153	#else
154	asm volatile(".byte 0x66, 0x0f, 0xae, 0xf1\t\n"
155		     :
156		     : "c"(ecx), "d"(edx), "a"(eax));
157	#endif
158}
159
160#endif /* _ASM_X86_MWAIT_H */
161