ARM: mvebu: extend PMSU code to support dynamic frequency scaling
authorThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Wed, 9 Jul 2014 15:45:10 +0000 (17:45 +0200)
committerJason Cooper <jason@lakedaemon.net>
Wed, 16 Jul 2014 12:58:36 +0000 (12:58 +0000)
This commit adds the necessary code in the Marvell EBU PMSU driver to
support dynamic frequency scaling. In essence, what this new code does
is that it:

 * registers the frequency operating points supported by the CPU;

 * registers a clock notifier of the CPU clocks. The notifier function
   listens to the newly introduced APPLY_RATE_CHANGE event, and uses
   that to finalize the frequency transition by doing the part of the
   procedure that involves the PMSU;

 * registers a platform device for the cpufreq-generic driver, which
   will take care of the CPU frequency transitions.

Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Link: https://lkml.kernel.org/r/1404920715-19834-3-git-send-email-thomas.petazzoni@free-electrons.com
Signed-off-by: Jason Cooper <jason@lakedaemon.net>
arch/arm/mach-mvebu/pmsu.c
include/linux/mvebu-pmsu.h [new file with mode: 0644]

index 53a55c8520bf9a1fbab97919457af791f202ca2d..db7d9ab298b64f7ad949a18918ef302b4d80928a 100644 (file)
 
 #define pr_fmt(fmt) "mvebu-pmsu: " fmt
 
+#include <linux/clk.h>
 #include <linux/cpu_pm.h>
+#include <linux/delay.h>
 #include <linux/kernel.h>
 #include <linux/init.h>
 #include <linux/of_address.h>
+#include <linux/of_device.h>
 #include <linux/io.h>
 #include <linux/platform_device.h>
+#include <linux/pm_opp.h>
 #include <linux/smp.h>
 #include <linux/resource.h>
+#include <linux/slab.h>
 #include <asm/cacheflush.h>
 #include <asm/cp15.h>
 #include <asm/smp_plat.h>
 #include <asm/suspend.h>
 #include <asm/tlbflush.h>
 #include "common.h"
+#include "armada-370-xp.h"
 
 static void __iomem *pmsu_mp_base;
 
@@ -57,6 +63,10 @@ static void __iomem *pmsu_mp_base;
 #define PMSU_STATUS_AND_MASK_IRQ_MASK          BIT(24)
 #define PMSU_STATUS_AND_MASK_FIQ_MASK          BIT(25)
 
+#define PMSU_EVENT_STATUS_AND_MASK(cpu)     ((cpu * 0x100) + 0x120)
+#define PMSU_EVENT_STATUS_AND_MASK_DFS_DONE        BIT(1)
+#define PMSU_EVENT_STATUS_AND_MASK_DFS_DONE_MASK   BIT(17)
+
 #define PMSU_BOOT_ADDR_REDIRECT_OFFSET(cpu) ((cpu * 0x100) + 0x124)
 
 /* PMSU fabric registers */
@@ -296,3 +306,155 @@ int __init armada_370_xp_cpu_pm_init(void)
 
 arch_initcall(armada_370_xp_cpu_pm_init);
 early_initcall(armada_370_xp_pmsu_init);
+
+static void mvebu_pmsu_dfs_request_local(void *data)
+{
+       u32 reg;
+       u32 cpu = smp_processor_id();
+       unsigned long flags;
+
+       local_irq_save(flags);
+
+       /* Prepare to enter idle */
+       reg = readl(pmsu_mp_base + PMSU_STATUS_AND_MASK(cpu));
+       reg |= PMSU_STATUS_AND_MASK_CPU_IDLE_WAIT |
+              PMSU_STATUS_AND_MASK_IRQ_MASK     |
+              PMSU_STATUS_AND_MASK_FIQ_MASK;
+       writel(reg, pmsu_mp_base + PMSU_STATUS_AND_MASK(cpu));
+
+       /* Request the DFS transition */
+       reg = readl(pmsu_mp_base + PMSU_CONTROL_AND_CONFIG(cpu));
+       reg |= PMSU_CONTROL_AND_CONFIG_DFS_REQ;
+       writel(reg, pmsu_mp_base + PMSU_CONTROL_AND_CONFIG(cpu));
+
+       /* The fact of entering idle will trigger the DFS transition */
+       wfi();
+
+       /*
+        * We're back from idle, the DFS transition has completed,
+        * clear the idle wait indication.
+        */
+       reg = readl(pmsu_mp_base + PMSU_STATUS_AND_MASK(cpu));
+       reg &= ~PMSU_STATUS_AND_MASK_CPU_IDLE_WAIT;
+       writel(reg, pmsu_mp_base + PMSU_STATUS_AND_MASK(cpu));
+
+       local_irq_restore(flags);
+}
+
+int mvebu_pmsu_dfs_request(int cpu)
+{
+       unsigned long timeout;
+       int hwcpu = cpu_logical_map(cpu);
+       u32 reg;
+
+       /* Clear any previous DFS DONE event */
+       reg = readl(pmsu_mp_base + PMSU_EVENT_STATUS_AND_MASK(hwcpu));
+       reg &= ~PMSU_EVENT_STATUS_AND_MASK_DFS_DONE;
+       writel(reg, pmsu_mp_base + PMSU_EVENT_STATUS_AND_MASK(hwcpu));
+
+       /* Mask the DFS done interrupt, since we are going to poll */
+       reg = readl(pmsu_mp_base + PMSU_EVENT_STATUS_AND_MASK(hwcpu));
+       reg |= PMSU_EVENT_STATUS_AND_MASK_DFS_DONE_MASK;
+       writel(reg, pmsu_mp_base + PMSU_EVENT_STATUS_AND_MASK(hwcpu));
+
+       /* Trigger the DFS on the appropriate CPU */
+       smp_call_function_single(cpu, mvebu_pmsu_dfs_request_local,
+                                NULL, false);
+
+       /* Poll until the DFS done event is generated */
+       timeout = jiffies + HZ;
+       while (time_before(jiffies, timeout)) {
+               reg = readl(pmsu_mp_base + PMSU_EVENT_STATUS_AND_MASK(hwcpu));
+               if (reg & PMSU_EVENT_STATUS_AND_MASK_DFS_DONE)
+                       break;
+               udelay(10);
+       }
+
+       if (time_after(jiffies, timeout))
+               return -ETIME;
+
+       /* Restore the DFS mask to its original state */
+       reg = readl(pmsu_mp_base + PMSU_EVENT_STATUS_AND_MASK(hwcpu));
+       reg &= ~PMSU_EVENT_STATUS_AND_MASK_DFS_DONE_MASK;
+       writel(reg, pmsu_mp_base + PMSU_EVENT_STATUS_AND_MASK(hwcpu));
+
+       return 0;
+}
+
+static int __init armada_xp_pmsu_cpufreq_init(void)
+{
+       struct device_node *np;
+       struct resource res;
+       int ret, cpu;
+
+       if (!of_machine_is_compatible("marvell,armadaxp"))
+               return 0;
+
+       /*
+        * In order to have proper cpufreq handling, we need to ensure
+        * that the Device Tree description of the CPU clock includes
+        * the definition of the PMU DFS registers. If not, we do not
+        * register the clock notifier and the cpufreq driver. This
+        * piece of code is only for compatibility with old Device
+        * Trees.
+        */
+       np = of_find_compatible_node(NULL, NULL, "marvell,armada-xp-cpu-clock");
+       if (!np)
+               return 0;
+
+       ret = of_address_to_resource(np, 1, &res);
+       if (ret) {
+               pr_warn(FW_WARN "not enabling cpufreq, deprecated armada-xp-cpu-clock binding\n");
+               of_node_put(np);
+               return 0;
+       }
+
+       of_node_put(np);
+
+       /*
+        * For each CPU, this loop registers the operating points
+        * supported (which are the nominal CPU frequency and half of
+        * it), and registers the clock notifier that will take care
+        * of doing the PMSU part of a frequency transition.
+        */
+       for_each_possible_cpu(cpu) {
+               struct device *cpu_dev;
+               struct clk *clk;
+               int ret;
+
+               cpu_dev = get_cpu_device(cpu);
+               if (!cpu_dev) {
+                       pr_err("Cannot get CPU %d\n", cpu);
+                       continue;
+               }
+
+               clk = clk_get(cpu_dev, 0);
+               if (!clk) {
+                       pr_err("Cannot get clock for CPU %d\n", cpu);
+                       return -ENODEV;
+               }
+
+               /*
+                * In case of a failure of dev_pm_opp_add(), we don't
+                * bother with cleaning up the registered OPP (there's
+                * no function to do so), and simply cancel the
+                * registration of the cpufreq device.
+                */
+               ret = dev_pm_opp_add(cpu_dev, clk_get_rate(clk), 0);
+               if (ret) {
+                       clk_put(clk);
+                       return ret;
+               }
+
+               ret = dev_pm_opp_add(cpu_dev, clk_get_rate(clk) / 2, 0);
+               if (ret) {
+                       clk_put(clk);
+                       return ret;
+               }
+       }
+
+       platform_device_register_simple("cpufreq-generic", -1, NULL, 0);
+       return 0;
+}
+
+device_initcall(armada_xp_pmsu_cpufreq_init);
diff --git a/include/linux/mvebu-pmsu.h b/include/linux/mvebu-pmsu.h
new file mode 100644 (file)
index 0000000..b918d07
--- /dev/null
@@ -0,0 +1,20 @@
+/*
+ * Copyright (C) 2012 Marvell
+ *
+ * Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
+ *
+ * This file is licensed under the terms of the GNU General Public
+ * License version 2.  This program is licensed "as is" without any
+ * warranty of any kind, whether express or implied.
+ */
+
+#ifndef __MVEBU_PMSU_H__
+#define __MVEBU_PMSU_H__
+
+#ifdef CONFIG_MACH_MVEBU_V7
+int mvebu_pmsu_dfs_request(int cpu);
+#else
+static inline int mvebu_pmsu_dfs_request(int cpu) { return -ENODEV; }
+#endif
+
+#endif /* __MVEBU_PMSU_H__ */