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, 2012 Cavium, Inc.
8  */
9 #include <linux/clocksource.h>
10 #include <linux/export.h>
11 #include <linux/init.h>
12 #include <linux/smp.h>
13 
14 #include <asm/cpu-info.h>
15 #include <asm/cpu-type.h>
16 #include <asm/time.h>
17 
18 #include <asm/octeon/octeon.h>
19 #include <asm/octeon/cvmx-ipd-defs.h>
20 #include <asm/octeon/cvmx-mio-defs.h>
21 
22 
23 static u64 f;
24 static u64 rdiv;
25 static u64 sdiv;
26 static u64 octeon_udelay_factor;
27 static u64 octeon_ndelay_factor;
28 
29 void __init octeon_setup_delays(void)
30 {
31 	octeon_udelay_factor = octeon_get_clock_rate() / 1000000;
32 	/*
33 	 * For __ndelay we divide by 2^16, so the factor is multiplied
34 	 * by the same amount.
35 	 */
36 	octeon_ndelay_factor = (octeon_udelay_factor * 0x10000ull) / 1000ull;
37 
38 	preset_lpj = octeon_get_clock_rate() / HZ;
39 
40 	if (current_cpu_type() == CPU_CAVIUM_OCTEON2) {
41 		union cvmx_mio_rst_boot rst_boot;
42 		rst_boot.u64 = cvmx_read_csr(CVMX_MIO_RST_BOOT);
43 		rdiv = rst_boot.s.c_mul;	/* CPU clock */
44 		sdiv = rst_boot.s.pnr_mul;	/* I/O clock */
45 		f = (0x8000000000000000ull / sdiv) * 2;
46 	}
47 }
48 
49 /*
50  * Set the current core's cvmcount counter to the value of the
51  * IPD_CLK_COUNT.  We do this on all cores as they are brought
52  * on-line.  This allows for a read from a local cpu register to
53  * access a synchronized counter.
54  *
55  * On CPU_CAVIUM_OCTEON2 the IPD_CLK_COUNT is scaled by rdiv/sdiv.
56  */
57 void octeon_init_cvmcount(void)
58 {
59 	unsigned long flags;
60 	unsigned loops = 2;
61 
62 	/* Clobber loops so GCC will not unroll the following while loop. */
63 	asm("" : "+r" (loops));
64 
65 	local_irq_save(flags);
66 	/*
67 	 * Loop several times so we are executing from the cache,
68 	 * which should give more deterministic timing.
69 	 */
70 	while (loops--) {
71 		u64 ipd_clk_count = cvmx_read_csr(CVMX_IPD_CLK_COUNT);
72 		if (rdiv != 0) {
73 			ipd_clk_count *= rdiv;
74 			if (f != 0) {
75 				asm("dmultu\t%[cnt],%[f]\n\t"
76 				    "mfhi\t%[cnt]"
77 				    : [cnt] "+r" (ipd_clk_count)
78 				    : [f] "r" (f)
79 				    : "hi", "lo");
80 			}
81 		}
82 		write_c0_cvmcount(ipd_clk_count);
83 	}
84 	local_irq_restore(flags);
85 }
86 
87 static cycle_t octeon_cvmcount_read(struct clocksource *cs)
88 {
89 	return read_c0_cvmcount();
90 }
91 
92 static struct clocksource clocksource_mips = {
93 	.name		= "OCTEON_CVMCOUNT",
94 	.read		= octeon_cvmcount_read,
95 	.mask		= CLOCKSOURCE_MASK(64),
96 	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
97 };
98 
99 unsigned long long notrace sched_clock(void)
100 {
101 	/* 64-bit arithmatic can overflow, so use 128-bit.  */
102 	u64 t1, t2, t3;
103 	unsigned long long rv;
104 	u64 mult = clocksource_mips.mult;
105 	u64 shift = clocksource_mips.shift;
106 	u64 cnt = read_c0_cvmcount();
107 
108 	asm (
109 		"dmultu\t%[cnt],%[mult]\n\t"
110 		"nor\t%[t1],$0,%[shift]\n\t"
111 		"mfhi\t%[t2]\n\t"
112 		"mflo\t%[t3]\n\t"
113 		"dsll\t%[t2],%[t2],1\n\t"
114 		"dsrlv\t%[rv],%[t3],%[shift]\n\t"
115 		"dsllv\t%[t1],%[t2],%[t1]\n\t"
116 		"or\t%[rv],%[t1],%[rv]\n\t"
117 		: [rv] "=&r" (rv), [t1] "=&r" (t1), [t2] "=&r" (t2), [t3] "=&r" (t3)
118 		: [cnt] "r" (cnt), [mult] "r" (mult), [shift] "r" (shift)
119 		: "hi", "lo");
120 	return rv;
121 }
122 
123 void __init plat_time_init(void)
124 {
125 	clocksource_mips.rating = 300;
126 	clocksource_register_hz(&clocksource_mips, octeon_get_clock_rate());
127 }
128 
129 void __udelay(unsigned long us)
130 {
131 	u64 cur, end, inc;
132 
133 	cur = read_c0_cvmcount();
134 
135 	inc = us * octeon_udelay_factor;
136 	end = cur + inc;
137 
138 	while (end > cur)
139 		cur = read_c0_cvmcount();
140 }
141 EXPORT_SYMBOL(__udelay);
142 
143 void __ndelay(unsigned long ns)
144 {
145 	u64 cur, end, inc;
146 
147 	cur = read_c0_cvmcount();
148 
149 	inc = ((ns * octeon_ndelay_factor) >> 16);
150 	end = cur + inc;
151 
152 	while (end > cur)
153 		cur = read_c0_cvmcount();
154 }
155 EXPORT_SYMBOL(__ndelay);
156 
157 void __delay(unsigned long loops)
158 {
159 	u64 cur, end;
160 
161 	cur = read_c0_cvmcount();
162 	end = cur + loops;
163 
164 	while (end > cur)
165 		cur = read_c0_cvmcount();
166 }
167 EXPORT_SYMBOL(__delay);
168 
169 
170 /**
171  * octeon_io_clk_delay - wait for a given number of io clock cycles to pass.
172  *
173  * We scale the wait by the clock ratio, and then wait for the
174  * corresponding number of core clocks.
175  *
176  * @count: The number of clocks to wait.
177  */
178 void octeon_io_clk_delay(unsigned long count)
179 {
180 	u64 cur, end;
181 
182 	cur = read_c0_cvmcount();
183 	if (rdiv != 0) {
184 		end = count * rdiv;
185 		if (f != 0) {
186 			asm("dmultu\t%[cnt],%[f]\n\t"
187 				"mfhi\t%[cnt]"
188 				: [cnt] "+r" (end)
189 				: [f] "r" (f)
190 				: "hi", "lo");
191 		}
192 		end = cur + end;
193 	} else {
194 		end = cur + count;
195 	}
196 	while (end > cur)
197 		cur = read_c0_cvmcount();
198 }
199 EXPORT_SYMBOL(octeon_io_clk_delay);
200