[Blackfin] arch: Functional power management support: Add support for cpu frequency...
Michael Hennerich [Thu, 24 Apr 2008 20:58:29 +0000 (04:58 +0800)]
Signed-off-by: Michael Hennerich <michael.hennerich@analog.com>
Signed-off-by: Bryan Wu <cooloney@kernel.org>

arch/blackfin/kernel/time-ts.c
arch/blackfin/kernel/time.c
arch/blackfin/mach-common/Makefile
arch/blackfin/mach-common/cpufreq.c [new file with mode: 0644]
include/asm-blackfin/time.h [new file with mode: 0644]

index 1ce8cb1..4482c47 100644 (file)
 #include <linux/irq.h>
 #include <linux/clocksource.h>
 #include <linux/clockchips.h>
+#include <linux/cpufreq.h>
 
 #include <asm/blackfin.h>
+#include <asm/time.h>
 
 #ifdef CONFIG_CYCLES_CLOCKSOURCE
 
+/* Accelerators for sched_clock()
+ * convert from cycles(64bits) => nanoseconds (64bits)
+ *  basic equation:
+ *             ns = cycles / (freq / ns_per_sec)
+ *             ns = cycles * (ns_per_sec / freq)
+ *             ns = cycles * (10^9 / (cpu_khz * 10^3))
+ *             ns = cycles * (10^6 / cpu_khz)
+ *
+ *     Then we use scaling math (suggested by george@mvista.com) to get:
+ *             ns = cycles * (10^6 * SC / cpu_khz) / SC
+ *             ns = cycles * cyc2ns_scale / SC
+ *
+ *     And since SC is a constant power of two, we can convert the div
+ *  into a shift.
+ *
+ *  We can use khz divisor instead of mhz to keep a better precision, since
+ *  cyc2ns_scale is limited to 10^6 * 2^10, which fits in 32 bits.
+ *  (mathieu.desnoyers@polymtl.ca)
+ *
+ *                     -johnstul@us.ibm.com "math is hard, lets go shopping!"
+ */
+
 static unsigned long cyc2ns_scale;
 #define CYC2NS_SCALE_FACTOR 10 /* 2^10, carefully chosen */
 
@@ -82,8 +106,9 @@ static void bfin_timer_set_mode(enum clock_event_mode mode,
 {
        switch (mode) {
        case CLOCK_EVT_MODE_PERIODIC: {
-               unsigned long tcount = ((get_cclk() / (HZ * 1)) - 1);
+               unsigned long tcount = ((get_cclk() / (HZ * TIME_SCALE)) - 1);
                bfin_write_TCNTL(TMPWR);
+               bfin_write_TSCALE(TIME_SCALE - 1);
                CSYNC();
                bfin_write_TPERIOD(tcount);
                bfin_write_TCOUNT(tcount);
@@ -92,6 +117,7 @@ static void bfin_timer_set_mode(enum clock_event_mode mode,
                break;
        }
        case CLOCK_EVT_MODE_ONESHOT:
+               bfin_write_TSCALE(0);
                bfin_write_TCOUNT(0);
                bfin_write_TCNTL(TMPWR | TMREN);
                CSYNC();
@@ -115,7 +141,7 @@ static void __init bfin_timer_init(void)
        /*
         * the TSCALE prescaler counter.
         */
-       bfin_write_TSCALE(0);
+       bfin_write_TSCALE(TIME_SCALE - 1);
        bfin_write_TPERIOD(0);
        bfin_write_TCOUNT(0);
 
index 715b394..eb23523 100644 (file)
@@ -6,9 +6,10 @@
  * Created:
  * Description:  This file contains the bfin-specific time handling details.
  *               Most of the stuff is located in the machine specific files.
+ *              FIXME: (This file is subject for removal)
  *
  * Modified:
- *               Copyright 2004-2006 Analog Devices Inc.
+ *               Copyright 2004-2008 Analog Devices Inc.
  *
  * Bugs:         Enter bugs at http://blackfin.uclinux.org/
  *
@@ -35,6 +36,7 @@
 #include <linux/irq.h>
 
 #include <asm/blackfin.h>
+#include <asm/time.h>
 
 /* This is an NTP setting */
 #define        TICK_SIZE (tick_nsec / 1000)
@@ -47,21 +49,6 @@ static struct irqaction bfin_timer_irq = {
        .flags = IRQF_DISABLED
 };
 
-/*
- * The way that the Blackfin core timer works is:
- *  - CCLK is divided by a programmable 8-bit pre-scaler (TSCALE)
- *  - Every time TSCALE ticks, a 32bit is counted down (TCOUNT)
- *
- * If you take the fastest clock (1ns, or 1GHz to make the math work easier)
- *    10ms is 10,000,000 clock ticks, which fits easy into a 32-bit counter
- *    (32 bit counter is 4,294,967,296ns or 4.2 seconds) so, we don't need
- *    to use TSCALE, and program it to zero (which is pass CCLK through).
- *    If you feel like using it, try to keep HZ * TIMESCALE to some
- *    value that divides easy (like power of 2).
- */
-
-#define TIME_SCALE 1
-
 static void
 time_sched_init(irq_handler_t timer_routine)
 {
index 15e33ca..393081e 100644 (file)
@@ -6,4 +6,5 @@ obj-y := \
        cache.o cacheinit.o entry.o \
        interrupt.o lock.o irqpanic.o arch_checks.o ints-priority.o
 
-obj-$(CONFIG_PM)                 += pm.o dpmc.o
+obj-$(CONFIG_PM)         += pm.o dpmc.o
+obj-$(CONFIG_CPU_FREQ)   += cpufreq.o
diff --git a/arch/blackfin/mach-common/cpufreq.c b/arch/blackfin/mach-common/cpufreq.c
new file mode 100644 (file)
index 0000000..ed81e00
--- /dev/null
@@ -0,0 +1,194 @@
+/*
+ * File:        arch/blackfin/mach-common/cpufreq.c
+ * Based on:
+ * Author:
+ *
+ * Created:
+ * Description:         Blackfin core clock scaling
+ *
+ * Modified:
+ *              Copyright 2004-2008 Analog Devices Inc.
+ *
+ * Bugs:        Enter bugs at http://blackfin.uclinux.org/
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.         See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, see the file COPYING, or write
+ * to the Free Software Foundation, Inc.,
+ * 51 Franklin St, Fifth Floor, Boston, MA     02110-1301      USA
+ */
+
+#include <linux/kernel.h>
+#include <linux/types.h>
+#include <linux/init.h>
+#include <linux/cpufreq.h>
+#include <linux/fs.h>
+#include <asm/blackfin.h>
+#include <asm/time.h>
+
+
+/* this is the table of CCLK frequencies, in Hz */
+/* .index is the entry in the auxillary dpm_state_table[] */
+static struct cpufreq_frequency_table bfin_freq_table[] = {
+       {
+               .frequency = CPUFREQ_TABLE_END,
+               .index = 0,
+       },
+       {
+               .frequency = CPUFREQ_TABLE_END,
+               .index = 1,
+       },
+       {
+               .frequency = CPUFREQ_TABLE_END,
+               .index = 2,
+       },
+       {
+               .frequency = CPUFREQ_TABLE_END,
+               .index = 0,
+       },
+};
+
+static struct bfin_dpm_state {
+       unsigned int csel; /* system clock divider */
+       unsigned int tscale; /* change the divider on the core timer interrupt */
+} dpm_state_table[3];
+
+/**************************************************************************/
+
+static unsigned int bfin_getfreq(unsigned int cpu)
+{
+       /* The driver only support single cpu */
+       if (cpu != 0)
+               return -1;
+
+       return get_cclk();
+}
+
+
+static int bfin_target(struct cpufreq_policy *policy,
+                       unsigned int target_freq, unsigned int relation)
+{
+       unsigned int index, plldiv, tscale;
+       unsigned long flags, cclk_hz;
+       struct cpufreq_freqs freqs;
+
+       if (cpufreq_frequency_table_target(policy, bfin_freq_table,
+                target_freq, relation, &index))
+               return -EINVAL;
+
+       cclk_hz = bfin_freq_table[index].frequency;
+
+       freqs.old = bfin_getfreq(0);
+       freqs.new = cclk_hz;
+       freqs.cpu = 0;
+
+       pr_debug("cpufreq: changing cclk to %lu; target = %u, oldfreq = %u\n",
+                cclk_hz, target_freq, freqs.old);
+
+       cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
+       local_irq_save(flags);
+               plldiv = (bfin_read_PLL_DIV() & SSEL) | dpm_state_table[index].csel;
+               tscale = dpm_state_table[index].tscale;
+               bfin_write_PLL_DIV(plldiv);
+               /* we have to adjust the core timer, because it is using cclk */
+               bfin_write_TSCALE(tscale);
+               SSYNC();
+       local_irq_restore(flags);
+       cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
+
+       return 0;
+}
+
+static int bfin_verify_speed(struct cpufreq_policy *policy)
+{
+       return cpufreq_frequency_table_verify(policy, bfin_freq_table);
+}
+
+static int __init __bfin_cpu_init(struct cpufreq_policy *policy)
+{
+
+       unsigned long cclk, sclk, csel, min_cclk;
+       int index;
+
+#ifdef CONFIG_CYCLES_CLOCKSOURCE
+/*
+ * Clocksource CYCLES is still CONTINUOUS but not longer MONOTONIC in case we enable
+ * CPU frequency scaling, since CYCLES runs off Core Clock.
+ */
+       printk(KERN_WARNING "CPU frequency scaling not supported: Clocksource not suitable\n"
+               return -ENODEV;
+#endif
+
+       if (policy->cpu != 0)
+               return -EINVAL;
+
+       cclk = get_cclk();
+       sclk = get_sclk();
+
+#if ANOMALY_05000273
+       min_cclk = sclk * 2;
+#else
+       min_cclk = sclk;
+#endif
+       csel = ((bfin_read_PLL_DIV() & CSEL) >> 4);
+
+       for (index = 0;  (cclk >> index) >= min_cclk && csel <= 3; index++, csel++) {
+               bfin_freq_table[index].frequency = cclk >> index;
+               dpm_state_table[index].csel = csel << 4; /* Shift now into PLL_DIV bitpos */
+               dpm_state_table[index].tscale =  (TIME_SCALE / (1 << csel)) - 1;
+
+               pr_debug("cpufreq: freq:%d csel:%d tscale:%d\n",
+                                                bfin_freq_table[index].frequency,
+                                                dpm_state_table[index].csel,
+                                                dpm_state_table[index].tscale);
+       }
+
+       policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
+
+       policy->cpuinfo.transition_latency = (bfin_read_PLL_LOCKCNT() / (sclk / 1000000)) * 1000;
+       /*Now ,only support one cpu */
+       policy->cur = cclk;
+       cpufreq_frequency_table_get_attr(bfin_freq_table, policy->cpu);
+       return cpufreq_frequency_table_cpuinfo(policy, bfin_freq_table);
+}
+
+static struct freq_attr *bfin_freq_attr[] = {
+       &cpufreq_freq_attr_scaling_available_freqs,
+       NULL,
+};
+
+static struct cpufreq_driver bfin_driver = {
+       .verify = bfin_verify_speed,
+       .target = bfin_target,
+       .get = bfin_getfreq,
+       .init = __bfin_cpu_init,
+       .name = "bfin cpufreq",
+       .owner = THIS_MODULE,
+       .attr = bfin_freq_attr,
+};
+
+static int __init bfin_cpu_init(void)
+{
+       return cpufreq_register_driver(&bfin_driver);
+}
+
+static void __exit bfin_cpu_exit(void)
+{
+       cpufreq_unregister_driver(&bfin_driver);
+}
+
+MODULE_AUTHOR("Michael Hennerich <hennerich@blackfin.uclinux.org>");
+MODULE_DESCRIPTION("cpufreq driver for Blackfin");
+MODULE_LICENSE("GPL");
+
+module_init(bfin_cpu_init);
+module_exit(bfin_cpu_exit);
diff --git a/include/asm-blackfin/time.h b/include/asm-blackfin/time.h
new file mode 100644 (file)
index 0000000..6e5859b
--- /dev/null
@@ -0,0 +1,36 @@
+/*
+ * asm-blackfin/time.h:
+ *
+ * Copyright 2004-2008 Analog Devices Inc.
+ *
+ * Licensed under the GPL-2 or later.
+ */
+
+#ifndef _ASM_BLACKFIN_TIME_H
+#define _ASM_BLACKFIN_TIME_H
+
+/*
+ * The way that the Blackfin core timer works is:
+ *  - CCLK is divided by a programmable 8-bit pre-scaler (TSCALE)
+ *  - Every time TSCALE ticks, a 32bit is counted down (TCOUNT)
+ *
+ * If you take the fastest clock (1ns, or 1GHz to make the math work easier)
+ *    10ms is 10,000,000 clock ticks, which fits easy into a 32-bit counter
+ *    (32 bit counter is 4,294,967,296ns or 4.2 seconds) so, we don't need
+ *    to use TSCALE, and program it to zero (which is pass CCLK through).
+ *    If you feel like using it, try to keep HZ * TIMESCALE to some
+ *    value that divides easy (like power of 2).
+ */
+
+#ifndef CONFIG_CPU_FREQ
+#define TIME_SCALE 1
+#else
+/*
+ * Blackfin CPU frequency scaling supports max Core Clock 1, 1/2 and 1/4 .
+ * Whenever we change the Core Clock frequency changes we immediately
+ * adjust the Core Timer Presale Register. This way we don't lose time.
+ */
+#define TIME_SCALE 4
+#endif
+
+#endif