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 
__monitor(const void * eax,unsigned long ecx,unsigned long edx)29 static __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 
__monitorx(const void * eax,unsigned long ecx,unsigned long edx)37 static __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 
__mwait(unsigned long eax,unsigned long ecx)45 static __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  */
__mwaitx(unsigned long eax,unsigned long ebx,unsigned long ecx)80 static __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 
__sti_mwait(unsigned long eax,unsigned long ecx)90 static __always_inline void __sti_mwait(unsigned long eax, unsigned long ecx)
91 {
92 	mds_idle_clear_cpu_buffers();
93 	/* "mwait %eax, %ecx;" */
94 	asm volatile("sti; .byte 0x0f, 0x01, 0xc9;"
95 		     :: "a" (eax), "c" (ecx));
96 }
97 
98 /*
99  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
100  * which can obviate IPI to trigger checking of need_resched.
101  * We execute MONITOR against need_resched and enter optimized wait state
102  * through MWAIT. Whenever someone changes need_resched, we would be woken
103  * up from MWAIT (without an IPI).
104  *
105  * New with Core Duo processors, MWAIT can take some hints based on CPU
106  * capability.
107  */
mwait_idle_with_hints(unsigned long eax,unsigned long ecx)108 static __always_inline void mwait_idle_with_hints(unsigned long eax, unsigned long ecx)
109 {
110 	if (static_cpu_has_bug(X86_BUG_MONITOR) || !current_set_polling_and_test()) {
111 		if (static_cpu_has_bug(X86_BUG_CLFLUSH_MONITOR)) {
112 			mb();
113 			clflush((void *)&current_thread_info()->flags);
114 			mb();
115 		}
116 
117 		__monitor((void *)&current_thread_info()->flags, 0, 0);
118 
119 		if (!need_resched()) {
120 			if (ecx & 1) {
121 				__mwait(eax, ecx);
122 			} else {
123 				__sti_mwait(eax, ecx);
124 				raw_local_irq_disable();
125 			}
126 		}
127 	}
128 	current_clr_polling();
129 }
130 
131 /*
132  * Caller can specify whether to enter C0.1 (low latency, less
133  * power saving) or C0.2 state (saves more power, but longer wakeup
134  * latency). This may be overridden by the IA32_UMWAIT_CONTROL MSR
135  * which can force requests for C0.2 to be downgraded to C0.1.
136  */
__tpause(u32 ecx,u32 edx,u32 eax)137 static inline void __tpause(u32 ecx, u32 edx, u32 eax)
138 {
139 	/* "tpause %ecx, %edx, %eax;" */
140 	#ifdef CONFIG_AS_TPAUSE
141 	asm volatile("tpause %%ecx\n"
142 		     :
143 		     : "c"(ecx), "d"(edx), "a"(eax));
144 	#else
145 	asm volatile(".byte 0x66, 0x0f, 0xae, 0xf1\t\n"
146 		     :
147 		     : "c"(ecx), "d"(edx), "a"(eax));
148 	#endif
149 }
150 
151 #endif /* _ASM_X86_MWAIT_H */
152