]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * Precise Delay Loops for i386 | |
4 | * | |
5 | * Copyright (C) 1993 Linus Torvalds | |
6 | * Copyright (C) 1997 Martin Mares <mj@atrey.karlin.mff.cuni.cz> | |
e01b70ef | 7 | * Copyright (C) 2008 Jiri Hladky <hladky _dot_ jiri _at_ gmail _dot_ com> |
1da177e4 LT |
8 | * |
9 | * The __delay function must _NOT_ be inlined as its execution time | |
10 | * depends wildly on alignment on many x86 processors. The additional | |
11 | * jump magic is needed to get the timing stable on all the CPU's | |
12 | * we have to worry about. | |
13 | */ | |
14 | ||
e683014c | 15 | #include <linux/export.h> |
1da177e4 | 16 | #include <linux/sched.h> |
941e492b | 17 | #include <linux/timex.h> |
35d5d08a | 18 | #include <linux/preempt.h> |
1da177e4 | 19 | #include <linux/delay.h> |
6f84fa2f | 20 | |
1da177e4 LT |
21 | #include <asm/processor.h> |
22 | #include <asm/delay.h> | |
23 | #include <asm/timer.h> | |
b466bdb6 | 24 | #include <asm/mwait.h> |
1da177e4 LT |
25 | |
26 | #ifdef CONFIG_SMP | |
6f84fa2f | 27 | # include <asm/smp.h> |
1da177e4 LT |
28 | #endif |
29 | ||
6f84fa2f JS |
30 | /* simple loop based delay: */ |
31 | static void delay_loop(unsigned long loops) | |
32 | { | |
f0fbf0ab | 33 | asm volatile( |
e01b70ef JH |
34 | " test %0,%0 \n" |
35 | " jz 3f \n" | |
36 | " jmp 1f \n" | |
37 | ||
38 | ".align 16 \n" | |
39 | "1: jmp 2f \n" | |
40 | ||
41 | ".align 16 \n" | |
ff1b15b6 | 42 | "2: dec %0 \n" |
e01b70ef | 43 | " jnz 2b \n" |
ff1b15b6 | 44 | "3: dec %0 \n" |
e01b70ef JH |
45 | |
46 | : /* we don't need output */ | |
47 | :"a" (loops) | |
48 | ); | |
6f84fa2f JS |
49 | } |
50 | ||
51 | /* TSC based delay: */ | |
a7f4255f | 52 | static void delay_tsc(unsigned long __loops) |
6f84fa2f | 53 | { |
9cfa1a02 | 54 | u64 bclock, now, loops = __loops; |
5c1ea082 | 55 | int cpu; |
6f84fa2f | 56 | |
5c1ea082 SR |
57 | preempt_disable(); |
58 | cpu = smp_processor_id(); | |
03b9730b | 59 | bclock = rdtsc_ordered(); |
5c1ea082 | 60 | for (;;) { |
03b9730b | 61 | now = rdtsc_ordered(); |
5c1ea082 SR |
62 | if ((now - bclock) >= loops) |
63 | break; | |
64 | ||
65 | /* Allow RT tasks to run */ | |
66 | preempt_enable(); | |
67 | rep_nop(); | |
68 | preempt_disable(); | |
69 | ||
70 | /* | |
71 | * It is possible that we moved to another CPU, and | |
72 | * since TSC's are per-cpu we need to calculate | |
73 | * that. The delay must guarantee that we wait "at | |
74 | * least" the amount of time. Being moved to another | |
75 | * CPU could make the wait longer but we just need to | |
76 | * make sure we waited long enough. Rebalance the | |
77 | * counter for this CPU. | |
78 | */ | |
79 | if (unlikely(cpu != smp_processor_id())) { | |
80 | loops -= (now - bclock); | |
81 | cpu = smp_processor_id(); | |
03b9730b | 82 | bclock = rdtsc_ordered(); |
5c1ea082 SR |
83 | } |
84 | } | |
35d5d08a | 85 | preempt_enable(); |
6f84fa2f JS |
86 | } |
87 | ||
b466bdb6 HR |
88 | /* |
89 | * On some AMD platforms, MWAITX has a configurable 32-bit timer, that | |
90 | * counts with TSC frequency. The input value is the loop of the | |
91 | * counter, it will exit when the timer expires. | |
92 | */ | |
93 | static void delay_mwaitx(unsigned long __loops) | |
94 | { | |
95 | u64 start, end, delay, loops = __loops; | |
96 | ||
88d879d2 JN |
97 | /* |
98 | * Timer value of 0 causes MWAITX to wait indefinitely, unless there | |
99 | * is a store on the memory monitored by MONITORX. | |
100 | */ | |
101 | if (loops == 0) | |
102 | return; | |
103 | ||
b466bdb6 HR |
104 | start = rdtsc_ordered(); |
105 | ||
106 | for (;;) { | |
107 | delay = min_t(u64, MWAITX_MAX_LOOPS, loops); | |
108 | ||
109 | /* | |
c482feef | 110 | * Use cpu_tss_rw as a cacheline-aligned, seldomly |
b466bdb6 HR |
111 | * accessed per-cpu variable as the monitor target. |
112 | */ | |
c482feef | 113 | __monitorx(raw_cpu_ptr(&cpu_tss_rw), 0, 0); |
b466bdb6 HR |
114 | |
115 | /* | |
116 | * AMD, like Intel, supports the EAX hint and EAX=0xf | |
117 | * means, do not enter any deep C-state and we use it | |
118 | * here in delay() to minimize wakeup latency. | |
119 | */ | |
120 | __mwaitx(MWAITX_DISABLE_CSTATES, delay, MWAITX_ECX_TIMER_ENABLE); | |
121 | ||
122 | end = rdtsc_ordered(); | |
123 | ||
124 | if (loops <= end - start) | |
125 | break; | |
126 | ||
127 | loops -= end - start; | |
128 | ||
129 | start = end; | |
130 | } | |
131 | } | |
132 | ||
6f84fa2f JS |
133 | /* |
134 | * Since we calibrate only once at boot, this | |
135 | * function should be set once at boot and not changed | |
136 | */ | |
137 | static void (*delay_fn)(unsigned long) = delay_loop; | |
138 | ||
139 | void use_tsc_delay(void) | |
140 | { | |
b466bdb6 HR |
141 | if (delay_fn == delay_loop) |
142 | delay_fn = delay_tsc; | |
143 | } | |
144 | ||
145 | void use_mwaitx_delay(void) | |
146 | { | |
147 | delay_fn = delay_mwaitx; | |
6f84fa2f JS |
148 | } |
149 | ||
a18e3690 | 150 | int read_current_timer(unsigned long *timer_val) |
6f84fa2f JS |
151 | { |
152 | if (delay_fn == delay_tsc) { | |
4ea1636b | 153 | *timer_val = rdtsc(); |
6f84fa2f JS |
154 | return 0; |
155 | } | |
156 | return -1; | |
157 | } | |
1da177e4 LT |
158 | |
159 | void __delay(unsigned long loops) | |
160 | { | |
6f84fa2f | 161 | delay_fn(loops); |
1da177e4 | 162 | } |
f0fbf0ab | 163 | EXPORT_SYMBOL(__delay); |
1da177e4 | 164 | |
81423c37 | 165 | noinline void __const_udelay(unsigned long xloops) |
1da177e4 | 166 | { |
4c45c516 | 167 | unsigned long lpj = this_cpu_read(cpu_info.loops_per_jiffy) ? : loops_per_jiffy; |
1da177e4 | 168 | int d0; |
6f84fa2f | 169 | |
1da177e4 | 170 | xloops *= 4; |
f0fbf0ab | 171 | asm("mull %%edx" |
1da177e4 | 172 | :"=d" (xloops), "=&a" (d0) |
4c45c516 | 173 | :"1" (xloops), "0" (lpj * (HZ / 4))); |
6f84fa2f JS |
174 | |
175 | __delay(++xloops); | |
1da177e4 | 176 | } |
f0fbf0ab | 177 | EXPORT_SYMBOL(__const_udelay); |
1da177e4 LT |
178 | |
179 | void __udelay(unsigned long usecs) | |
180 | { | |
6f84fa2f | 181 | __const_udelay(usecs * 0x000010c7); /* 2**32 / 1000000 (rounded up) */ |
1da177e4 | 182 | } |
f0fbf0ab | 183 | EXPORT_SYMBOL(__udelay); |
1da177e4 LT |
184 | |
185 | void __ndelay(unsigned long nsecs) | |
186 | { | |
6f84fa2f | 187 | __const_udelay(nsecs * 0x00005); /* 2**32 / 1000000000 (rounded up) */ |
1da177e4 | 188 | } |
129f6946 | 189 | EXPORT_SYMBOL(__ndelay); |