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
7  * Copyright (C) 2009, 2010 Cavium Networks, Inc.
8  */
9 #include <linux/clocksource.h>
10 #include <linux/init.h>
11 #include <linux/smp.h>
12 
13 #include <asm/cpu-info.h>
14 #include <asm/time.h>
15 
16 #include <asm/octeon/octeon.h>
17 #include <asm/octeon/cvmx-ipd-defs.h>
18 #include <asm/octeon/cvmx-mio-defs.h>
19 
20 /*
21  * Set the current core's cvmcount counter to the value of the
22  * IPD_CLK_COUNT.  We do this on all cores as they are brought
23  * on-line.  This allows for a read from a local cpu register to
24  * access a synchronized counter.
25  *
26  * On CPU_CAVIUM_OCTEON2 the IPD_CLK_COUNT is scaled by rdiv/sdiv.
27  */
28 void octeon_init_cvmcount(void)
29 {
30 	unsigned long flags;
31 	unsigned loops = 2;
32 	u64 f = 0;
33 	u64 rdiv = 0;
34 	u64 sdiv = 0;
35 	if (current_cpu_type() == CPU_CAVIUM_OCTEON2) {
36 		union cvmx_mio_rst_boot rst_boot;
37 		rst_boot.u64 = cvmx_read_csr(CVMX_MIO_RST_BOOT);
38 		rdiv = rst_boot.s.c_mul;	/* CPU clock */
39 		sdiv = rst_boot.s.pnr_mul;	/* I/O clock */
40 		f = (0x8000000000000000ull / sdiv) * 2;
41 	}
42 
43 
44 	/* Clobber loops so GCC will not unroll the following while loop. */
45 	asm("" : "+r" (loops));
46 
47 	local_irq_save(flags);
48 	/*
49 	 * Loop several times so we are executing from the cache,
50 	 * which should give more deterministic timing.
51 	 */
52 	while (loops--) {
53 		u64 ipd_clk_count = cvmx_read_csr(CVMX_IPD_CLK_COUNT);
54 		if (rdiv != 0) {
55 			ipd_clk_count *= rdiv;
56 			if (f != 0) {
57 				asm("dmultu\t%[cnt],%[f]\n\t"
58 				    "mfhi\t%[cnt]"
59 				    : [cnt] "+r" (ipd_clk_count),
60 				      [f] "=r" (f)
61 				    : : "hi", "lo");
62 			}
63 		}
64 		write_c0_cvmcount(ipd_clk_count);
65 	}
66 	local_irq_restore(flags);
67 }
68 
69 static cycle_t octeon_cvmcount_read(struct clocksource *cs)
70 {
71 	return read_c0_cvmcount();
72 }
73 
74 static struct clocksource clocksource_mips = {
75 	.name		= "OCTEON_CVMCOUNT",
76 	.read		= octeon_cvmcount_read,
77 	.mask		= CLOCKSOURCE_MASK(64),
78 	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
79 };
80 
81 unsigned long long notrace sched_clock(void)
82 {
83 	/* 64-bit arithmatic can overflow, so use 128-bit.  */
84 	u64 t1, t2, t3;
85 	unsigned long long rv;
86 	u64 mult = clocksource_mips.mult;
87 	u64 shift = clocksource_mips.shift;
88 	u64 cnt = read_c0_cvmcount();
89 
90 	asm (
91 		"dmultu\t%[cnt],%[mult]\n\t"
92 		"nor\t%[t1],$0,%[shift]\n\t"
93 		"mfhi\t%[t2]\n\t"
94 		"mflo\t%[t3]\n\t"
95 		"dsll\t%[t2],%[t2],1\n\t"
96 		"dsrlv\t%[rv],%[t3],%[shift]\n\t"
97 		"dsllv\t%[t1],%[t2],%[t1]\n\t"
98 		"or\t%[rv],%[t1],%[rv]\n\t"
99 		: [rv] "=&r" (rv), [t1] "=&r" (t1), [t2] "=&r" (t2), [t3] "=&r" (t3)
100 		: [cnt] "r" (cnt), [mult] "r" (mult), [shift] "r" (shift)
101 		: "hi", "lo");
102 	return rv;
103 }
104 
105 void __init plat_time_init(void)
106 {
107 	clocksource_mips.rating = 300;
108 	clocksource_set_clock(&clocksource_mips, octeon_get_clock_rate());
109 	clocksource_register(&clocksource_mips);
110 }
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);
166