Commit | Line | Data |
---|---|---|
5b3b1688 DD |
1 | /* |
2 | * This file is subject to the terms and conditions of the GNU General Public | |
3 | * License. See the file "COPYING" in the main directory of this archive | |
4 | * for more details. | |
5 | * | |
6 | * Copyright (C) 2007 by Ralf Baechle | |
54954a6d | 7 | * Copyright (C) 2009, 2010 Cavium Networks, Inc. |
5b3b1688 DD |
8 | */ |
9 | #include <linux/clocksource.h> | |
cae39d13 | 10 | #include <linux/export.h> |
5b3b1688 | 11 | #include <linux/init.h> |
54954a6d | 12 | #include <linux/smp.h> |
5b3b1688 | 13 | |
54954a6d | 14 | #include <asm/cpu-info.h> |
5b3b1688 DD |
15 | #include <asm/time.h> |
16 | ||
17 | #include <asm/octeon/octeon.h> | |
18 | #include <asm/octeon/cvmx-ipd-defs.h> | |
54954a6d | 19 | #include <asm/octeon/cvmx-mio-defs.h> |
5b3b1688 DD |
20 | |
21 | /* | |
22 | * Set the current core's cvmcount counter to the value of the | |
23 | * IPD_CLK_COUNT. We do this on all cores as they are brought | |
24 | * on-line. This allows for a read from a local cpu register to | |
25 | * access a synchronized counter. | |
26 | * | |
54954a6d | 27 | * On CPU_CAVIUM_OCTEON2 the IPD_CLK_COUNT is scaled by rdiv/sdiv. |
5b3b1688 DD |
28 | */ |
29 | void octeon_init_cvmcount(void) | |
30 | { | |
31 | unsigned long flags; | |
32 | unsigned loops = 2; | |
54954a6d DD |
33 | u64 f = 0; |
34 | u64 rdiv = 0; | |
35 | u64 sdiv = 0; | |
36 | if (current_cpu_type() == CPU_CAVIUM_OCTEON2) { | |
37 | union cvmx_mio_rst_boot rst_boot; | |
38 | rst_boot.u64 = cvmx_read_csr(CVMX_MIO_RST_BOOT); | |
39 | rdiv = rst_boot.s.c_mul; /* CPU clock */ | |
40 | sdiv = rst_boot.s.pnr_mul; /* I/O clock */ | |
41 | f = (0x8000000000000000ull / sdiv) * 2; | |
42 | } | |
43 | ||
5b3b1688 DD |
44 | |
45 | /* Clobber loops so GCC will not unroll the following while loop. */ | |
46 | asm("" : "+r" (loops)); | |
47 | ||
48 | local_irq_save(flags); | |
49 | /* | |
50 | * Loop several times so we are executing from the cache, | |
51 | * which should give more deterministic timing. | |
52 | */ | |
54954a6d DD |
53 | while (loops--) { |
54 | u64 ipd_clk_count = cvmx_read_csr(CVMX_IPD_CLK_COUNT); | |
55 | if (rdiv != 0) { | |
56 | ipd_clk_count *= rdiv; | |
57 | if (f != 0) { | |
58 | asm("dmultu\t%[cnt],%[f]\n\t" | |
59 | "mfhi\t%[cnt]" | |
60 | : [cnt] "+r" (ipd_clk_count), | |
61 | [f] "=r" (f) | |
62 | : : "hi", "lo"); | |
63 | } | |
64 | } | |
65 | write_c0_cvmcount(ipd_clk_count); | |
66 | } | |
5b3b1688 DD |
67 | local_irq_restore(flags); |
68 | } | |
69 | ||
d0ce9a5a | 70 | static cycle_t octeon_cvmcount_read(struct clocksource *cs) |
5b3b1688 DD |
71 | { |
72 | return read_c0_cvmcount(); | |
73 | } | |
74 | ||
75 | static struct clocksource clocksource_mips = { | |
76 | .name = "OCTEON_CVMCOUNT", | |
77 | .read = octeon_cvmcount_read, | |
78 | .mask = CLOCKSOURCE_MASK(64), | |
79 | .flags = CLOCK_SOURCE_IS_CONTINUOUS, | |
80 | }; | |
81 | ||
c6a3c851 DD |
82 | unsigned long long notrace sched_clock(void) |
83 | { | |
0e8a1d82 | 84 | /* 64-bit arithmatic can overflow, so use 128-bit. */ |
0e8a1d82 DD |
85 | u64 t1, t2, t3; |
86 | unsigned long long rv; | |
87 | u64 mult = clocksource_mips.mult; | |
88 | u64 shift = clocksource_mips.shift; | |
89 | u64 cnt = read_c0_cvmcount(); | |
90 | ||
91 | asm ( | |
92 | "dmultu\t%[cnt],%[mult]\n\t" | |
93 | "nor\t%[t1],$0,%[shift]\n\t" | |
94 | "mfhi\t%[t2]\n\t" | |
95 | "mflo\t%[t3]\n\t" | |
96 | "dsll\t%[t2],%[t2],1\n\t" | |
97 | "dsrlv\t%[rv],%[t3],%[shift]\n\t" | |
98 | "dsllv\t%[t1],%[t2],%[t1]\n\t" | |
99 | "or\t%[rv],%[t1],%[rv]\n\t" | |
100 | : [rv] "=&r" (rv), [t1] "=&r" (t1), [t2] "=&r" (t2), [t3] "=&r" (t3) | |
101 | : [cnt] "r" (cnt), [mult] "r" (mult), [shift] "r" (shift) | |
102 | : "hi", "lo"); | |
103 | return rv; | |
c6a3c851 DD |
104 | } |
105 | ||
5b3b1688 DD |
106 | void __init plat_time_init(void) |
107 | { | |
108 | clocksource_mips.rating = 300; | |
75c4fd8c | 109 | clocksource_register_hz(&clocksource_mips, octeon_get_clock_rate()); |
5b3b1688 | 110 | } |
ca148125 DD |
111 | |
112 | static u64 octeon_udelay_factor; | |
113 | static u64 octeon_ndelay_factor; | |
114 | ||
115 | void __init octeon_setup_delays(void) | |
116 | { | |
117 | octeon_udelay_factor = octeon_get_clock_rate() / 1000000; | |
118 | /* | |
119 | * For __ndelay we divide by 2^16, so the factor is multiplied | |
120 | * by the same amount. | |
121 | */ | |
122 | octeon_ndelay_factor = (octeon_udelay_factor * 0x10000ull) / 1000ull; | |
123 | ||
124 | preset_lpj = octeon_get_clock_rate() / HZ; | |
125 | } | |
126 | ||
127 | void __udelay(unsigned long us) | |
128 | { | |
129 | u64 cur, end, inc; | |
130 | ||
131 | cur = read_c0_cvmcount(); | |
132 | ||
133 | inc = us * octeon_udelay_factor; | |
134 | end = cur + inc; | |
135 | ||
136 | while (end > cur) | |
137 | cur = read_c0_cvmcount(); | |
138 | } | |
139 | EXPORT_SYMBOL(__udelay); | |
140 | ||
141 | void __ndelay(unsigned long ns) | |
142 | { | |
143 | u64 cur, end, inc; | |
144 | ||
145 | cur = read_c0_cvmcount(); | |
146 | ||
147 | inc = ((ns * octeon_ndelay_factor) >> 16); | |
148 | end = cur + inc; | |
149 | ||
150 | while (end > cur) | |
151 | cur = read_c0_cvmcount(); | |
152 | } | |
153 | EXPORT_SYMBOL(__ndelay); | |
154 | ||
155 | void __delay(unsigned long loops) | |
156 | { | |
157 | u64 cur, end; | |
158 | ||
159 | cur = read_c0_cvmcount(); | |
160 | end = cur + loops; | |
161 | ||
162 | while (end > cur) | |
163 | cur = read_c0_cvmcount(); | |
164 | } | |
165 | EXPORT_SYMBOL(__delay); |