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