[CPUFREQ][2/2] preregister support for powernow-k8

This patch provides support for the _PSD ACPI object in the Powernow-k8
driver.  Although it looks like an invasive patch, most of it is
simply the consequence of turning the static acpi_performance_data
structure into a pointer.

AMD has tested it on several machines over the past few days without issue.

[trivial checkpatch warnings fixed up by davej]
[X86_POWERNOW_K8_ACPI=n buildfix from Randy Dunlap]

Signed-off-by: Mark Langsdorf <mark.langsdorf@amd.com>
Tested-by: Frank Arnold <frank.arnold@amd.com>
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Dave Jones <davej@redhat.com>

diff --git a/arch/x86/kernel/cpu/cpufreq/powernow-k8.c b/arch/x86/kernel/cpu/cpufreq/powernow-k8.c
index 84bb395038d879823d5beb4e4bf2e181b8bd6b91..4e7271999a74e58ccf15203d005a646d9135686a 100644 (file)
--- a/arch/x86/kernel/cpu/cpufreq/powernow-k8.c
+++ b/arch/x86/kernel/cpu/cpufreq/powernow-k8.c
@@ -737,44 +737,63 @@ static int find_psb_table(struct powernow_k8_data *data)
 #ifdef CONFIG_X86_POWERNOW_K8_ACPI
 static void powernow_k8_acpi_pst_values(struct powernow_k8_data *data, unsigned int index)
 {
-       if (!data->acpi_data.state_count || (cpu_family == CPU_HW_PSTATE))
+       if (!data->acpi_data->state_count || (cpu_family == CPU_HW_PSTATE))
                return;
 
-       data->irt = (data->acpi_data.states[index].control >> IRT_SHIFT) & IRT_MASK;
-       data->rvo = (data->acpi_data.states[index].control >> RVO_SHIFT) & RVO_MASK;
-       data->exttype = (data->acpi_data.states[index].control >> EXT_TYPE_SHIFT) & EXT_TYPE_MASK;
-       data->plllock = (data->acpi_data.states[index].control >> PLL_L_SHIFT) & PLL_L_MASK;
-       data->vidmvs = 1 << ((data->acpi_data.states[index].control >> MVS_SHIFT) & MVS_MASK);
-       data->vstable = (data->acpi_data.states[index].control >> VST_SHIFT) & VST_MASK;
+       data->irt = (data->acpi_data->states[index].control >> IRT_SHIFT) & IRT_MASK;
+       data->rvo = (data->acpi_data->states[index].control >> RVO_SHIFT) & RVO_MASK;
+       data->exttype = (data->acpi_data->states[index].control >> EXT_TYPE_SHIFT) & EXT_TYPE_MASK;
+       data->plllock = (data->acpi_data->states[index].control >> PLL_L_SHIFT) & PLL_L_MASK;
+       data->vidmvs = 1 << ((data->acpi_data->states[index].control >> MVS_SHIFT) & MVS_MASK);
+       data->vstable = (data->acpi_data->states[index].control >> VST_SHIFT) & VST_MASK;
+}
+
+
+static struct acpi_processor_performance *acpi_perf_data;
+static int preregister_valid;
+
+static int powernow_k8_cpu_preinit_acpi(void)
+{
+       acpi_perf_data = alloc_percpu(struct acpi_processor_performance);
+       if (!acpi_perf_data)
+               return -ENODEV;
+
+       if (acpi_processor_preregister_performance(acpi_perf_data))
+               return -ENODEV;
+       else
+               preregister_valid = 1;
+       return 0;
 }
 
 static int powernow_k8_cpu_init_acpi(struct powernow_k8_data *data)
 {
        struct cpufreq_frequency_table *powernow_table;
        int ret_val;
+       int cpu = 0;
 
-       if (acpi_processor_register_performance(&data->acpi_data, data->cpu)) {
+       data->acpi_data = percpu_ptr(acpi_perf_data, cpu);
+       if (acpi_processor_register_performance(data->acpi_data, data->cpu)) {
                dprintk("register performance failed: bad ACPI data\n");
                return -EIO;
        }
 
        /* verify the data contained in the ACPI structures */
-       if (data->acpi_data.state_count <= 1) {
+       if (data->acpi_data->state_count <= 1) {
                dprintk("No ACPI P-States\n");
                goto err_out;
        }
 
-       if ((data->acpi_data.control_register.space_id != ACPI_ADR_SPACE_FIXED_HARDWARE) ||
-               (data->acpi_data.status_register.space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)) {
+       if ((data->acpi_data->control_register.space_id != ACPI_ADR_SPACE_FIXED_HARDWARE) ||
+               (data->acpi_data->status_register.space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)) {
                dprintk("Invalid control/status registers (%x - %x)\n",
-                       data->acpi_data.control_register.space_id,
-                       data->acpi_data.status_register.space_id);
+                       data->acpi_data->control_register.space_id,
+                       data->acpi_data->status_register.space_id);
                goto err_out;
        }
 
        /* fill in data->powernow_table */
        powernow_table = kmalloc((sizeof(struct cpufreq_frequency_table)
-               * (data->acpi_data.state_count + 1)), GFP_KERNEL);
+               * (data->acpi_data->state_count + 1)), GFP_KERNEL);
        if (!powernow_table) {
                dprintk("powernow_table memory alloc failure\n");
                goto err_out;
@@ -787,12 +806,12 @@ static int powernow_k8_cpu_init_acpi(struct powernow_k8_data *data)
        if (ret_val)
                goto err_out_mem;
 
-       powernow_table[data->acpi_data.state_count].frequency = CPUFREQ_TABLE_END;
-       powernow_table[data->acpi_data.state_count].index = 0;
+       powernow_table[data->acpi_data->state_count].frequency = CPUFREQ_TABLE_END;
+       powernow_table[data->acpi_data->state_count].index = 0;
        data->powernow_table = powernow_table;
 
        /* fill in data */
-       data->numps = data->acpi_data.state_count;
+       data->numps = data->acpi_data->state_count;
        if (first_cpu(per_cpu(cpu_core_map, data->cpu)) == data->cpu)
                print_basics(data);
        powernow_k8_acpi_pst_values(data, 0);
@@ -800,16 +819,31 @@ static int powernow_k8_cpu_init_acpi(struct powernow_k8_data *data)
        /* notify BIOS that we exist */
        acpi_processor_notify_smm(THIS_MODULE);
 
+       /* determine affinity, from ACPI if available */
+       if (preregister_valid) {
+               if ((data->acpi_data->shared_type == CPUFREQ_SHARED_TYPE_ALL) ||
+                       (data->acpi_data->shared_type == CPUFREQ_SHARED_TYPE_ANY))
+                       data->starting_core_affinity = data->acpi_data->shared_cpu_map;
+               else
+                       data->starting_core_affinity = cpumask_of_cpu(data->cpu);
+       } else {
+       /* best guess from family if not */
+               if (cpu_family == CPU_HW_PSTATE)
+                       data->starting_core_affinity = cpumask_of_cpu(data->cpu);
+               else
+                       data->starting_core_affinity = per_cpu(cpu_core_map, data->cpu);
+       }
+
        return 0;
 
 err_out_mem:
        kfree(powernow_table);
 
 err_out:
-       acpi_processor_unregister_performance(&data->acpi_data, data->cpu);
+       acpi_processor_unregister_performance(data->acpi_data, data->cpu);
 
        /* data->acpi_data.state_count informs us at ->exit() whether ACPI was used */
-       data->acpi_data.state_count = 0;
+       data->acpi_data->state_count = 0;
 
        return -ENODEV;
 }
@@ -821,10 +855,10 @@ static int fill_powernow_table_pstate(struct powernow_k8_data *data, struct cpuf
        rdmsr(MSR_PSTATE_CUR_LIMIT, hi, lo);
        data->max_hw_pstate = (hi & HW_PSTATE_MAX_MASK) >> HW_PSTATE_MAX_SHIFT;
 
-       for (i = 0; i < data->acpi_data.state_count; i++) {
+       for (i = 0; i < data->acpi_data->state_count; i++) {
                u32 index;
 
-               index = data->acpi_data.states[i].control & HW_PSTATE_MASK;
+               index = data->acpi_data->states[i].control & HW_PSTATE_MASK;
                if (index > data->max_hw_pstate) {
                        printk(KERN_ERR PFX "invalid pstate %d - bad value %d.\n", i, index);
                        printk(KERN_ERR PFX "Please report to BIOS manufacturer\n");
@@ -840,7 +874,7 @@ static int fill_powernow_table_pstate(struct powernow_k8_data *data, struct cpuf
 
                powernow_table[i].index = index;
 
-               powernow_table[i].frequency = data->acpi_data.states[i].core_frequency * 1000;
+               powernow_table[i].frequency = data->acpi_data->states[i].core_frequency * 1000;
        }
        return 0;
 }
@@ -849,16 +883,16 @@ static int fill_powernow_table_fidvid(struct powernow_k8_data *data, struct cpuf
 {
        int i;
        int cntlofreq = 0;
-       for (i = 0; i < data->acpi_data.state_count; i++) {
+       for (i = 0; i < data->acpi_data->state_count; i++) {
                u32 fid;
                u32 vid;
 
                if (data->exttype) {
-                       fid = data->acpi_data.states[i].status & EXT_FID_MASK;
-                       vid = (data->acpi_data.states[i].status >> VID_SHIFT) & EXT_VID_MASK;
+                       fid = data->acpi_data->states[i].status & EXT_FID_MASK;
+                       vid = (data->acpi_data->states[i].status >> VID_SHIFT) & EXT_VID_MASK;
                } else {
-                       fid = data->acpi_data.states[i].control & FID_MASK;
-                       vid = (data->acpi_data.states[i].control >> VID_SHIFT) & VID_MASK;
+                       fid = data->acpi_data->states[i].control & FID_MASK;
+                       vid = (data->acpi_data->states[i].control >> VID_SHIFT) & VID_MASK;
                }
 
                dprintk("   %d : fid 0x%x, vid 0x%x\n", i, fid, vid);
@@ -899,10 +933,10 @@ static int fill_powernow_table_fidvid(struct powernow_k8_data *data, struct cpuf
                                cntlofreq = i;
                }
 
-               if (powernow_table[i].frequency != (data->acpi_data.states[i].core_frequency * 1000)) {
+               if (powernow_table[i].frequency != (data->acpi_data->states[i].core_frequency * 1000)) {
                        printk(KERN_INFO PFX "invalid freq entries %u kHz vs. %u kHz\n",
                                powernow_table[i].frequency,
-                               (unsigned int) (data->acpi_data.states[i].core_frequency * 1000));
+                               (unsigned int) (data->acpi_data->states[i].core_frequency * 1000));
                        powernow_table[i].frequency = CPUFREQ_ENTRY_INVALID;
                        continue;
                }
@@ -912,11 +946,12 @@ static int fill_powernow_table_fidvid(struct powernow_k8_data *data, struct cpuf
 
 static void powernow_k8_cpu_exit_acpi(struct powernow_k8_data *data)
 {
-       if (data->acpi_data.state_count)
-               acpi_processor_unregister_performance(&data->acpi_data, data->cpu);
+       if (data->acpi_data->state_count)
+               acpi_processor_unregister_performance(data->acpi_data, data->cpu);
 }
 
 #else
+static int powernow_k8_cpu_preinit_acpi(void) { return -ENODEV; }
 static int powernow_k8_cpu_init_acpi(struct powernow_k8_data *data) { return -ENODEV; }
 static void powernow_k8_cpu_exit_acpi(struct powernow_k8_data *data) { return; }
 static void powernow_k8_acpi_pst_values(struct powernow_k8_data *data, unsigned int index) { return; }
@@ -1101,7 +1136,7 @@ static int powernowk8_verify(struct cpufreq_policy *pol)
 static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol)
 {
        struct powernow_k8_data *data;
-       cpumask_t oldmask;
+       cpumask_t oldmask = CPU_MASK_ALL;
        int rc;
 
        if (!cpu_online(pol->cpu))
@@ -1174,10 +1209,7 @@ static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol)
        /* run on any CPU again */
        set_cpus_allowed_ptr(current, &oldmask);
 
-       if (cpu_family == CPU_HW_PSTATE)
-               pol->cpus = cpumask_of_cpu(pol->cpu);
-       else
-               pol->cpus = per_cpu(cpu_core_map, pol->cpu);
+       pol->cpus = data->starting_core_affinity;
        data->available_cores = &(pol->cpus);
 
        /* Take a crude guess here.
@@ -1300,6 +1332,7 @@ static int __cpuinit powernowk8_init(void)
        }
 
        if (supported_cpus == num_online_cpus()) {
+               powernow_k8_cpu_preinit_acpi();
                printk(KERN_INFO PFX "Found %d %s "
                        "processors (%d cpu cores) (" VERSION ")\n",
                        num_online_nodes(),
@@ -1316,6 +1349,10 @@ static void __exit powernowk8_exit(void)
        dprintk("exit\n");
 
        cpufreq_unregister_driver(&cpufreq_amd64_driver);
+
+#ifdef CONFIG_X86_POWERNOW_K8_ACPI
+       free_percpu(acpi_perf_data);
+#endif
 }
 
 MODULE_AUTHOR("Paul Devriendt <paul.devriendt@amd.com> and Mark Langsdorf <mark.langsdorf@amd.com>");

diff --git a/arch/x86/kernel/cpu/cpufreq/powernow-k8.h b/arch/x86/kernel/cpu/cpufreq/powernow-k8.h
index ab48cfed4d96ce5629e29f3a21ffd9c3e25d8a21..a62612cd4be89c2e775edf5d9c71da41cf0a4b64 100644 (file)
--- a/arch/x86/kernel/cpu/cpufreq/powernow-k8.h
+++ b/arch/x86/kernel/cpu/cpufreq/powernow-k8.h
@@ -33,12 +33,13 @@ struct powernow_k8_data {
 #ifdef CONFIG_X86_POWERNOW_K8_ACPI
        /* the acpi table needs to be kept. it's only available if ACPI was
         * used to determine valid frequency/vid/fid states */
-       struct acpi_processor_performance acpi_data;
+       struct acpi_processor_performance *acpi_data;
 #endif
        /* we need to keep track of associated cores, but let cpufreq
         * handle hotplug events - so just point at cpufreq pol->cpus
         * structure */
        cpumask_t *available_cores;
+       cpumask_t starting_core_affinity;
 };