These changes are the raw update to linux-4.4.6-rt14. Kernel sources
[kvmfornfv.git] / kernel / drivers / hwmon / fam15h_power.c
index 3057dfc..5f7067d 100644 (file)
@@ -26,6 +26,7 @@
 #include <linux/pci.h>
 #include <linux/bitops.h>
 #include <asm/processor.h>
+#include <asm/msr.h>
 
 MODULE_DESCRIPTION("AMD Family 15h CPU processor power monitor");
 MODULE_AUTHOR("Andreas Herrmann <herrmann.der.user@googlemail.com>");
@@ -41,11 +42,21 @@ MODULE_LICENSE("GPL");
 #define REG_TDP_RUNNING_AVERAGE                0xe0
 #define REG_TDP_LIMIT3                 0xe8
 
+#define FAM15H_MIN_NUM_ATTRS           2
+#define FAM15H_NUM_GROUPS              2
+
+#define MSR_F15H_CU_MAX_PWR_ACCUMULATOR        0xc001007b
+
 struct fam15h_power_data {
        struct pci_dev *pdev;
        unsigned int tdp_to_watts;
        unsigned int base_tdp;
        unsigned int processor_pwr_watts;
+       unsigned int cpu_pwr_sample_ratio;
+       const struct attribute_group *groups[FAM15H_NUM_GROUPS];
+       struct attribute_group group;
+       /* maximum accumulated power of a compute unit */
+       u64 max_cu_acc_power;
 };
 
 static ssize_t show_power(struct device *dev,
@@ -59,8 +70,19 @@ static ssize_t show_power(struct device *dev,
 
        pci_bus_read_config_dword(f4->bus, PCI_DEVFN(PCI_SLOT(f4->devfn), 5),
                                  REG_TDP_RUNNING_AVERAGE, &val);
-       running_avg_capture = (val >> 4) & 0x3fffff;
-       running_avg_capture = sign_extend32(running_avg_capture, 21);
+
+       /*
+        * On Carrizo and later platforms, TdpRunAvgAccCap bit field
+        * is extended to 4:31 from 4:25.
+        */
+       if (boot_cpu_data.x86 == 0x15 && boot_cpu_data.x86_model >= 0x60) {
+               running_avg_capture = val >> 4;
+               running_avg_capture = sign_extend32(running_avg_capture, 27);
+       } else {
+               running_avg_capture = (val >> 4) & 0x3fffff;
+               running_avg_capture = sign_extend32(running_avg_capture, 21);
+       }
+
        running_avg_range = (val & 0xf) + 1;
 
        pci_bus_read_config_dword(f4->bus, PCI_DEVFN(PCI_SLOT(f4->devfn), 5),
@@ -93,31 +115,38 @@ static ssize_t show_power_crit(struct device *dev,
 }
 static DEVICE_ATTR(power1_crit, S_IRUGO, show_power_crit, NULL);
 
-static umode_t fam15h_power_is_visible(struct kobject *kobj,
-                                      struct attribute *attr,
-                                      int index)
+static int fam15h_power_init_attrs(struct pci_dev *pdev,
+                                  struct fam15h_power_data *data)
 {
-       /* power1_input is only reported for Fam15h, Models 00h-0fh */
-       if (attr == &dev_attr_power1_input.attr &&
-          (boot_cpu_data.x86 != 0x15 || boot_cpu_data.x86_model > 0xf))
-               return 0;
+       int n = FAM15H_MIN_NUM_ATTRS;
+       struct attribute **fam15h_power_attrs;
+       struct cpuinfo_x86 *c = &boot_cpu_data;
 
-       return attr->mode;
-}
+       if (c->x86 == 0x15 &&
+           (c->x86_model <= 0xf ||
+            (c->x86_model >= 0x60 && c->x86_model <= 0x6f)))
+               n += 1;
 
-static struct attribute *fam15h_power_attrs[] = {
-       &dev_attr_power1_input.attr,
-       &dev_attr_power1_crit.attr,
-       NULL
-};
+       fam15h_power_attrs = devm_kcalloc(&pdev->dev, n,
+                                         sizeof(*fam15h_power_attrs),
+                                         GFP_KERNEL);
 
-static const struct attribute_group fam15h_power_group = {
-       .attrs = fam15h_power_attrs,
-       .is_visible = fam15h_power_is_visible,
-};
-__ATTRIBUTE_GROUPS(fam15h_power);
+       if (!fam15h_power_attrs)
+               return -ENOMEM;
+
+       n = 0;
+       fam15h_power_attrs[n++] = &dev_attr_power1_crit.attr;
+       if (c->x86 == 0x15 &&
+           (c->x86_model <= 0xf ||
+            (c->x86_model >= 0x60 && c->x86_model <= 0x6f)))
+               fam15h_power_attrs[n++] = &dev_attr_power1_input.attr;
+
+       data->group.attrs = fam15h_power_attrs;
+
+       return 0;
+}
 
-static bool fam15h_power_is_internal_node0(struct pci_dev *f4)
+static bool should_load_on_this_node(struct pci_dev *f4)
 {
        u32 val;
 
@@ -174,11 +203,12 @@ static int fam15h_power_resume(struct pci_dev *pdev)
 #define fam15h_power_resume NULL
 #endif
 
-static void fam15h_power_init_data(struct pci_dev *f4,
-                                            struct fam15h_power_data *data)
+static int fam15h_power_init_data(struct pci_dev *f4,
+                                 struct fam15h_power_data *data)
 {
-       u32 val;
+       u32 val, eax, ebx, ecx, edx;
        u64 tmp;
+       int ret;
 
        pci_read_config_dword(f4, REG_PROCESSOR_TDP, &val);
        data->base_tdp = val >> 16;
@@ -198,14 +228,41 @@ static void fam15h_power_init_data(struct pci_dev *f4,
 
        /* convert to microWatt */
        data->processor_pwr_watts = (tmp * 15625) >> 10;
+
+       ret = fam15h_power_init_attrs(f4, data);
+       if (ret)
+               return ret;
+
+       cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
+
+       /* CPUID Fn8000_0007:EDX[12] indicates to support accumulated power */
+       if (!(edx & BIT(12)))
+               return 0;
+
+       /*
+        * determine the ratio of the compute unit power accumulator
+        * sample period to the PTSC counter period by executing CPUID
+        * Fn8000_0007:ECX
+        */
+       data->cpu_pwr_sample_ratio = ecx;
+
+       if (rdmsrl_safe(MSR_F15H_CU_MAX_PWR_ACCUMULATOR, &tmp)) {
+               pr_err("Failed to read max compute unit power accumulator MSR\n");
+               return -ENODEV;
+       }
+
+       data->max_cu_acc_power = tmp;
+
+       return 0;
 }
 
 static int fam15h_power_probe(struct pci_dev *pdev,
-                                       const struct pci_device_id *id)
+                             const struct pci_device_id *id)
 {
        struct fam15h_power_data *data;
        struct device *dev = &pdev->dev;
        struct device *hwmon_dev;
+       int ret;
 
        /*
         * though we ignore every other northbridge, we still have to
@@ -214,25 +271,31 @@ static int fam15h_power_probe(struct pci_dev *pdev,
         */
        tweak_runavg_range(pdev);
 
-       if (!fam15h_power_is_internal_node0(pdev))
+       if (!should_load_on_this_node(pdev))
                return -ENODEV;
 
        data = devm_kzalloc(dev, sizeof(struct fam15h_power_data), GFP_KERNEL);
        if (!data)
                return -ENOMEM;
 
-       fam15h_power_init_data(pdev, data);
+       ret = fam15h_power_init_data(pdev, data);
+       if (ret)
+               return ret;
+
        data->pdev = pdev;
 
+       data->groups[0] = &data->group;
+
        hwmon_dev = devm_hwmon_device_register_with_groups(dev, "fam15h_power",
                                                           data,
-                                                          fam15h_power_groups);
+                                                          &data->groups[0]);
        return PTR_ERR_OR_ZERO(hwmon_dev);
 }
 
 static const struct pci_device_id fam15h_power_id_table[] = {
        { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_NB_F4) },
        { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_M30H_NB_F4) },
+       { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_M60H_NB_F4) },
        { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_16H_NB_F4) },
        { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_16H_M30H_NB_F4) },
        {}