This commit was manufactured by cvs2svn to create branch 'vserver'.
[linux-2.6.git] / drivers / acpi / processor_perflib.c
diff --git a/drivers/acpi/processor_perflib.c b/drivers/acpi/processor_perflib.c
new file mode 100644 (file)
index 0000000..a9a1a8f
--- /dev/null
@@ -0,0 +1,666 @@
+/*
+ * processor_perflib.c - ACPI Processor P-States Library ($Revision: 71 $)
+ *
+ *  Copyright (C) 2001, 2002 Andy Grover <andrew.grover@intel.com>
+ *  Copyright (C) 2001, 2002 Paul Diefenbaugh <paul.s.diefenbaugh@intel.com>
+ *  Copyright (C) 2004       Dominik Brodowski <linux@brodo.de>
+ *  Copyright (C) 2004  Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
+ *                     - Added processor hotplug support
+ *
+ *
+ * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ *
+ *  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, write to the Free Software Foundation, Inc.,
+ *  59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.
+ *
+ */
+
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/cpufreq.h>
+
+#ifdef CONFIG_X86_ACPI_CPUFREQ_PROC_INTF
+#include <linux/proc_fs.h>
+#include <linux/seq_file.h>
+
+#include <asm/uaccess.h>
+#endif
+
+#include <acpi/acpi_bus.h>
+#include <acpi/processor.h>
+
+
+#define ACPI_PROCESSOR_COMPONENT       0x01000000
+#define ACPI_PROCESSOR_CLASS           "processor"
+#define ACPI_PROCESSOR_DRIVER_NAME     "ACPI Processor Driver"
+#define ACPI_PROCESSOR_FILE_PERFORMANCE        "performance"
+#define _COMPONENT             ACPI_PROCESSOR_COMPONENT
+ACPI_MODULE_NAME               ("acpi_processor")
+
+
+static DECLARE_MUTEX(performance_sem);
+
+/*
+ * _PPC support is implemented as a CPUfreq policy notifier:
+ * This means each time a CPUfreq driver registered also with
+ * the ACPI core is asked to change the speed policy, the maximum
+ * value is adjusted so that it is within the platform limit.
+ *
+ * Also, when a new platform limit value is detected, the CPUfreq
+ * policy is adjusted accordingly.
+ */
+
+#define PPC_REGISTERED   1
+#define PPC_IN_USE       2
+
+static int acpi_processor_ppc_status = 0;
+
+static int acpi_processor_ppc_notifier(struct notifier_block *nb,
+       unsigned long event,
+       void *data)
+{
+       struct cpufreq_policy *policy = data;
+       struct acpi_processor *pr;
+       unsigned int ppc = 0;
+
+       down(&performance_sem);
+
+       if (event != CPUFREQ_INCOMPATIBLE)
+               goto out;
+
+       pr = processors[policy->cpu];
+       if (!pr || !pr->performance)
+               goto out;
+
+       ppc = (unsigned int) pr->performance_platform_limit;
+       if (!ppc)
+               goto out;
+
+       if (ppc > pr->performance->state_count)
+               goto out;
+
+       cpufreq_verify_within_limits(policy, 0,
+               pr->performance->states[ppc].core_frequency * 1000);
+
+ out:
+       up(&performance_sem);
+
+       return 0;
+}
+
+
+static struct notifier_block acpi_ppc_notifier_block = {
+       .notifier_call = acpi_processor_ppc_notifier,
+};
+
+
+static int
+acpi_processor_get_platform_limit (
+       struct acpi_processor*  pr)
+{
+       acpi_status             status = 0;
+       unsigned long           ppc = 0;
+
+       ACPI_FUNCTION_TRACE("acpi_processor_get_platform_limit");
+
+       if (!pr)
+               return_VALUE(-EINVAL);
+
+       /*
+        * _PPC indicates the maximum state currently supported by the platform
+        * (e.g. 0 = states 0..n; 1 = states 1..n; etc.
+        */
+       status = acpi_evaluate_integer(pr->handle, "_PPC", NULL, &ppc);
+
+       if (status != AE_NOT_FOUND)
+               acpi_processor_ppc_status |= PPC_IN_USE;
+
+       if(ACPI_FAILURE(status) && status != AE_NOT_FOUND) {
+               ACPI_DEBUG_PRINT((ACPI_DB_ERROR, "Error evaluating _PPC\n"));
+               return_VALUE(-ENODEV);
+       }
+
+       pr->performance_platform_limit = (int) ppc;
+
+       return_VALUE(0);
+}
+
+
+int acpi_processor_ppc_has_changed(
+       struct acpi_processor *pr)
+{
+       int ret = acpi_processor_get_platform_limit(pr);
+       if (ret < 0)
+               return (ret);
+       else
+               return cpufreq_update_policy(pr->id);
+}
+
+
+void acpi_processor_ppc_init(void) {
+       if (!cpufreq_register_notifier(&acpi_ppc_notifier_block, CPUFREQ_POLICY_NOTIFIER))
+               acpi_processor_ppc_status |= PPC_REGISTERED;
+       else
+               printk(KERN_DEBUG "Warning: Processor Platform Limit not supported.\n");
+}
+
+
+void acpi_processor_ppc_exit(void) {
+       if (acpi_processor_ppc_status & PPC_REGISTERED)
+               cpufreq_unregister_notifier(&acpi_ppc_notifier_block, CPUFREQ_POLICY_NOTIFIER);
+
+       acpi_processor_ppc_status &= ~PPC_REGISTERED;
+}
+
+/*
+ * when registering a cpufreq driver with this ACPI processor driver, the
+ * _PCT and _PSS structures are read out and written into struct
+ * acpi_processor_performance.
+ */
+static int acpi_processor_set_pdc (struct acpi_processor *pr)
+{
+       acpi_status             status = AE_OK;
+       u32                     arg0_buf[3];
+       union acpi_object       arg0 = {ACPI_TYPE_BUFFER};
+       struct acpi_object_list no_object = {1, &arg0};
+       struct acpi_object_list *pdc;
+
+       ACPI_FUNCTION_TRACE("acpi_processor_set_pdc");
+
+       arg0.buffer.length = 12;
+       arg0.buffer.pointer = (u8 *) arg0_buf;
+       arg0_buf[0] = ACPI_PDC_REVISION_ID;
+       arg0_buf[1] = 0;
+       arg0_buf[2] = 0;
+
+       pdc = (pr->performance->pdc) ? pr->performance->pdc : &no_object;
+
+       status = acpi_evaluate_object(pr->handle, "_PDC", pdc, NULL);
+
+       if ((ACPI_FAILURE(status)) && (pr->performance->pdc))
+               ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Error evaluating _PDC, using legacy perf. control...\n"));
+
+       return_VALUE(status);
+}
+
+
+static int
+acpi_processor_get_performance_control (
+       struct acpi_processor *pr)
+{
+       int                     result = 0;
+       acpi_status             status = 0;
+       struct acpi_buffer      buffer = {ACPI_ALLOCATE_BUFFER, NULL};
+       union acpi_object       *pct = NULL;
+       union acpi_object       obj = {0};
+
+       ACPI_FUNCTION_TRACE("acpi_processor_get_performance_control");
+
+       status = acpi_evaluate_object(pr->handle, "_PCT", NULL, &buffer);
+       if(ACPI_FAILURE(status)) {
+               ACPI_DEBUG_PRINT((ACPI_DB_ERROR, "Error evaluating _PCT\n"));
+               return_VALUE(-ENODEV);
+       }
+
+       pct = (union acpi_object *) buffer.pointer;
+       if (!pct || (pct->type != ACPI_TYPE_PACKAGE)
+               || (pct->package.count != 2)) {
+               ACPI_DEBUG_PRINT((ACPI_DB_ERROR, "Invalid _PCT data\n"));
+               result = -EFAULT;
+               goto end;
+       }
+
+       /*
+        * control_register
+        */
+
+       obj = pct->package.elements[0];
+
+       if ((obj.type != ACPI_TYPE_BUFFER)
+               || (obj.buffer.length < sizeof(struct acpi_pct_register))
+               || (obj.buffer.pointer == NULL)) {
+               ACPI_DEBUG_PRINT((ACPI_DB_ERROR,
+                       "Invalid _PCT data (control_register)\n"));
+               result = -EFAULT;
+               goto end;
+       }
+       memcpy(&pr->performance->control_register, obj.buffer.pointer, sizeof(struct acpi_pct_register));
+
+
+       /*
+        * status_register
+        */
+
+       obj = pct->package.elements[1];
+
+       if ((obj.type != ACPI_TYPE_BUFFER)
+               || (obj.buffer.length < sizeof(struct acpi_pct_register))
+               || (obj.buffer.pointer == NULL)) {
+               ACPI_DEBUG_PRINT((ACPI_DB_ERROR,
+                       "Invalid _PCT data (status_register)\n"));
+               result = -EFAULT;
+               goto end;
+       }
+
+       memcpy(&pr->performance->status_register, obj.buffer.pointer, sizeof(struct acpi_pct_register));
+
+end:
+       acpi_os_free(buffer.pointer);
+
+       return_VALUE(result);
+}
+
+
+static int
+acpi_processor_get_performance_states (
+       struct acpi_processor   *pr)
+{
+       int                     result = 0;
+       acpi_status             status = AE_OK;
+       struct acpi_buffer      buffer = {ACPI_ALLOCATE_BUFFER, NULL};
+       struct acpi_buffer      format = {sizeof("NNNNNN"), "NNNNNN"};
+       struct acpi_buffer      state = {0, NULL};
+       union acpi_object       *pss = NULL;
+       int                     i;
+
+       ACPI_FUNCTION_TRACE("acpi_processor_get_performance_states");
+
+       status = acpi_evaluate_object(pr->handle, "_PSS", NULL, &buffer);
+       if(ACPI_FAILURE(status)) {
+               ACPI_DEBUG_PRINT((ACPI_DB_ERROR, "Error evaluating _PSS\n"));
+               return_VALUE(-ENODEV);
+       }
+
+       pss = (union acpi_object *) buffer.pointer;
+       if (!pss || (pss->type != ACPI_TYPE_PACKAGE)) {
+               ACPI_DEBUG_PRINT((ACPI_DB_ERROR, "Invalid _PSS data\n"));
+               result = -EFAULT;
+               goto end;
+       }
+
+       ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d performance states\n",
+               pss->package.count));
+
+       pr->performance->state_count = pss->package.count;
+       pr->performance->states = kmalloc(sizeof(struct acpi_processor_px) * pss->package.count, GFP_KERNEL);
+       if (!pr->performance->states) {
+               result = -ENOMEM;
+               goto end;
+       }
+
+       for (i = 0; i < pr->performance->state_count; i++) {
+
+               struct acpi_processor_px *px = &(pr->performance->states[i]);
+
+               state.length = sizeof(struct acpi_processor_px);
+               state.pointer = px;
+
+               ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Extracting state %d\n", i));
+
+               status = acpi_extract_package(&(pss->package.elements[i]),
+                       &format, &state);
+               if (ACPI_FAILURE(status)) {
+                       ACPI_DEBUG_PRINT((ACPI_DB_ERROR, "Invalid _PSS data\n"));
+                       result = -EFAULT;
+                       kfree(pr->performance->states);
+                       goto end;
+               }
+
+               ACPI_DEBUG_PRINT((ACPI_DB_INFO,
+                       "State [%d]: core_frequency[%d] power[%d] transition_latency[%d] bus_master_latency[%d] control[0x%x] status[0x%x]\n",
+                       i,
+                       (u32) px->core_frequency,
+                       (u32) px->power,
+                       (u32) px->transition_latency,
+                       (u32) px->bus_master_latency,
+                       (u32) px->control,
+                       (u32) px->status));
+
+               if (!px->core_frequency) {
+                       ACPI_DEBUG_PRINT((ACPI_DB_ERROR, "Invalid _PSS data: freq is zero\n"));
+                       result = -EFAULT;
+                       kfree(pr->performance->states);
+                       goto end;
+               }
+       }
+
+end:
+       acpi_os_free(buffer.pointer);
+
+       return_VALUE(result);
+}
+
+
+static int
+acpi_processor_get_performance_info (
+       struct acpi_processor   *pr)
+{
+       int                     result = 0;
+       acpi_status             status = AE_OK;
+       acpi_handle             handle = NULL;
+
+       ACPI_FUNCTION_TRACE("acpi_processor_get_performance_info");
+
+       if (!pr || !pr->performance || !pr->handle)
+               return_VALUE(-EINVAL);
+
+       acpi_processor_set_pdc(pr);
+
+       status = acpi_get_handle(pr->handle, "_PCT", &handle);
+       if (ACPI_FAILURE(status)) {
+               ACPI_DEBUG_PRINT((ACPI_DB_INFO,
+                       "ACPI-based processor performance control unavailable\n"));
+               return_VALUE(-ENODEV);
+       }
+
+       result = acpi_processor_get_performance_control(pr);
+       if (result)
+               return_VALUE(result);
+
+       result = acpi_processor_get_performance_states(pr);
+       if (result)
+               return_VALUE(result);
+
+       result = acpi_processor_get_platform_limit(pr);
+       if (result)
+               return_VALUE(result);
+
+       return_VALUE(0);
+}
+
+
+int acpi_processor_notify_smm(struct module *calling_module) {
+       acpi_status             status;
+       static int              is_done = 0;
+
+       ACPI_FUNCTION_TRACE("acpi_processor_notify_smm");
+
+       if (!(acpi_processor_ppc_status & PPC_REGISTERED))
+               return_VALUE(-EBUSY);
+
+       if (!try_module_get(calling_module))
+               return_VALUE(-EINVAL);
+
+       /* is_done is set to negative if an error occured,
+        * and to postitive if _no_ error occured, but SMM
+        * was already notified. This avoids double notification
+        * which might lead to unexpected results...
+        */
+       if (is_done > 0) {
+               module_put(calling_module);
+               return_VALUE(0);
+       }
+       else if (is_done < 0) {
+               module_put(calling_module);
+               return_VALUE(is_done);
+       }
+
+       is_done = -EIO;
+
+       /* Can't write pstate_cnt to smi_cmd if either value is zero */
+       if ((!acpi_fadt.smi_cmd) ||
+           (!acpi_fadt.pstate_cnt)) {
+               ACPI_DEBUG_PRINT((ACPI_DB_INFO,
+                       "No SMI port or pstate_cnt\n"));
+               module_put(calling_module);
+               return_VALUE(0);
+       }
+
+       ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Writing pstate_cnt [0x%x] to smi_cmd [0x%x]\n", acpi_fadt.pstate_cnt, acpi_fadt.smi_cmd));
+
+       /* FADT v1 doesn't support pstate_cnt, many BIOS vendors use
+        * it anyway, so we need to support it... */
+       if (acpi_fadt_is_v1) {
+               ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Using v1.0 FADT reserved value for pstate_cnt\n"));
+       }
+
+       status = acpi_os_write_port (acpi_fadt.smi_cmd,
+                                    (u32) acpi_fadt.pstate_cnt, 8);
+       if (ACPI_FAILURE (status)) {
+               ACPI_DEBUG_PRINT((ACPI_DB_ERROR,
+                                 "Failed to write pstate_cnt [0x%x] to "
+                                 "smi_cmd [0x%x]\n", acpi_fadt.pstate_cnt, acpi_fadt.smi_cmd));
+               module_put(calling_module);
+               return_VALUE(status);
+       }
+
+       /* Success. If there's no _PPC, we need to fear nothing, so
+        * we can allow the cpufreq driver to be rmmod'ed. */
+       is_done = 1;
+
+       if (!(acpi_processor_ppc_status & PPC_IN_USE))
+               module_put(calling_module);
+
+       return_VALUE(0);
+}
+EXPORT_SYMBOL(acpi_processor_notify_smm);
+
+
+#ifdef CONFIG_X86_ACPI_CPUFREQ_PROC_INTF
+/* /proc/acpi/processor/../performance interface (DEPRECATED) */
+
+static int acpi_processor_perf_open_fs(struct inode *inode, struct file *file);
+static struct file_operations acpi_processor_perf_fops = {
+       .open           = acpi_processor_perf_open_fs,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = single_release,
+};
+
+static int acpi_processor_perf_seq_show(struct seq_file *seq, void *offset)
+{
+       struct acpi_processor   *pr = (struct acpi_processor *)seq->private;
+       int                     i;
+
+       ACPI_FUNCTION_TRACE("acpi_processor_perf_seq_show");
+
+       if (!pr)
+               goto end;
+
+       if (!pr->performance) {
+               seq_puts(seq, "<not supported>\n");
+               goto end;
+       }
+
+       seq_printf(seq, "state count:             %d\n"
+                       "active state:            P%d\n",
+                       pr->performance->state_count,
+                       pr->performance->state);
+
+       seq_puts(seq, "states:\n");
+       for (i = 0; i < pr->performance->state_count; i++)
+               seq_printf(seq, "   %cP%d:                  %d MHz, %d mW, %d uS\n",
+                       (i == pr->performance->state?'*':' '), i,
+                       (u32) pr->performance->states[i].core_frequency,
+                       (u32) pr->performance->states[i].power,
+                       (u32) pr->performance->states[i].transition_latency);
+
+end:
+       return_VALUE(0);
+}
+
+static int acpi_processor_perf_open_fs(struct inode *inode, struct file *file)
+{
+       return single_open(file, acpi_processor_perf_seq_show,
+                                               PDE(inode)->data);
+}
+
+static ssize_t
+acpi_processor_write_performance (
+        struct file            *file,
+        const char             __user *buffer,
+        size_t                 count,
+        loff_t                 *data)
+{
+       int                     result = 0;
+       struct seq_file         *m = (struct seq_file *) file->private_data;
+       struct acpi_processor   *pr = (struct acpi_processor *) m->private;
+       struct acpi_processor_performance *perf;
+       char                    state_string[12] = {'\0'};
+       unsigned int            new_state = 0;
+       struct cpufreq_policy   policy;
+
+       ACPI_FUNCTION_TRACE("acpi_processor_write_performance");
+
+       if (!pr || (count > sizeof(state_string) - 1))
+               return_VALUE(-EINVAL);
+
+       perf = pr->performance;
+       if (!perf)
+               return_VALUE(-EINVAL);
+
+       if (copy_from_user(state_string, buffer, count))
+               return_VALUE(-EFAULT);
+
+       state_string[count] = '\0';
+       new_state = simple_strtoul(state_string, NULL, 0);
+
+       if (new_state >= perf->state_count)
+               return_VALUE(-EINVAL);
+
+       cpufreq_get_policy(&policy, pr->id);
+
+       policy.cpu = pr->id;
+       policy.min = perf->states[new_state].core_frequency * 1000;
+       policy.max = perf->states[new_state].core_frequency * 1000;
+
+       result = cpufreq_set_policy(&policy);
+       if (result)
+               return_VALUE(result);
+
+       return_VALUE(count);
+}
+
+static void
+acpi_cpufreq_add_file (
+       struct acpi_processor *pr)
+{
+       struct proc_dir_entry   *entry = NULL;
+       struct acpi_device      *device = NULL;
+
+       ACPI_FUNCTION_TRACE("acpi_cpufreq_addfile");
+
+       if (acpi_bus_get_device(pr->handle, &device))
+               return_VOID;
+
+       /* add file 'performance' [R/W] */
+       entry = create_proc_entry(ACPI_PROCESSOR_FILE_PERFORMANCE,
+                 S_IFREG|S_IRUGO|S_IWUSR, acpi_device_dir(device));
+       if (!entry)
+               ACPI_DEBUG_PRINT((ACPI_DB_ERROR,
+                       "Unable to create '%s' fs entry\n",
+                       ACPI_PROCESSOR_FILE_PERFORMANCE));
+       else {
+               entry->proc_fops = &acpi_processor_perf_fops;
+               entry->proc_fops->write = acpi_processor_write_performance;
+               entry->data = acpi_driver_data(device);
+               entry->owner = THIS_MODULE;
+       }
+       return_VOID;
+}
+
+static void
+acpi_cpufreq_remove_file (
+       struct acpi_processor *pr)
+{
+       struct acpi_device      *device = NULL;
+
+       ACPI_FUNCTION_TRACE("acpi_cpufreq_addfile");
+
+       if (acpi_bus_get_device(pr->handle, &device))
+               return_VOID;
+
+       /* remove file 'performance' */
+       remove_proc_entry(ACPI_PROCESSOR_FILE_PERFORMANCE,
+                 acpi_device_dir(device));
+
+       return_VOID;
+}
+
+#else
+static void acpi_cpufreq_add_file (struct acpi_processor *pr) { return; }
+static void acpi_cpufreq_remove_file (struct acpi_processor *pr) { return; }
+#endif /* CONFIG_X86_ACPI_CPUFREQ_PROC_INTF */
+
+
+int
+acpi_processor_register_performance (
+       struct acpi_processor_performance * performance,
+       unsigned int cpu)
+{
+       struct acpi_processor *pr;
+
+       ACPI_FUNCTION_TRACE("acpi_processor_register_performance");
+
+       if (!(acpi_processor_ppc_status & PPC_REGISTERED))
+               return_VALUE(-EINVAL);
+
+       down(&performance_sem);
+
+       pr = processors[cpu];
+       if (!pr) {
+               up(&performance_sem);
+               return_VALUE(-ENODEV);
+       }
+
+       if (pr->performance) {
+               up(&performance_sem);
+               return_VALUE(-EBUSY);
+       }
+
+       pr->performance = performance;
+
+       if (acpi_processor_get_performance_info(pr)) {
+               pr->performance = NULL;
+               up(&performance_sem);
+               return_VALUE(-EIO);
+       }
+
+       acpi_cpufreq_add_file(pr);
+
+       up(&performance_sem);
+       return_VALUE(0);
+}
+EXPORT_SYMBOL(acpi_processor_register_performance);
+
+
+void
+acpi_processor_unregister_performance (
+       struct acpi_processor_performance * performance,
+       unsigned int cpu)
+{
+       struct acpi_processor *pr;
+
+       ACPI_FUNCTION_TRACE("acpi_processor_unregister_performance");
+
+       down(&performance_sem);
+
+       pr = processors[cpu];
+       if (!pr) {
+               up(&performance_sem);
+               return_VOID;
+       }
+
+       kfree(pr->performance->states);
+       pr->performance = NULL;
+
+       acpi_cpufreq_remove_file(pr);
+
+       up(&performance_sem);
+
+       return_VOID;
+}
+EXPORT_SYMBOL(acpi_processor_unregister_performance);