On 04/08/2015 03:26 PM, Guenter Roeck wrote:
> On 04/08/2015 12:00 AM, Cédric Le Goater wrote:
>> The new OPAL device tree adds a few properties which can be used to add
>> extra information on the sensor label.
>>
>> In the case of a cpu core sensor, the firmware exposes the physical
>> identifier of the core in the "ibm,pir" property. The driver
>> translates this identifier in a linux cpu number and prints out a
>> range corresponding to the hardware threads of the core (as they
>> share the same sensor).
>>
>> The numbering gives a hint on the localization of the core in the
>> system (which socket, which chip).
>>
>> Signed-off-by: Cédric Le Goater <clg@xxxxxxxxxx>
>> ---
>>
>> Changes since v2:
>>
>> - fix bogus logical cpu retrieval
>> - use 'threads_per_core' to print out cpu range
>>
>> Changes since v1:
>>
>> - check cpu validity before printing out the attribute label.
>> if invalid, use a "phy" prefix to distinguish a linux cpu
>> number from a physical cpu number.
>>
>> drivers/hwmon/ibmpowernv.c | 43 +++++++++++++++++++++++++++++++++++++++++++
>> 1 file changed, 43 insertions(+)
>>
>> Index: linux.git/drivers/hwmon/ibmpowernv.c
>> ===================================================================
>> --- linux.git.orig/drivers/hwmon/ibmpowernv.c
>> +++ linux.git/drivers/hwmon/ibmpowernv.c
>> @@ -30,6 +30,7 @@
>> #include <linux/platform_device.h>
>> #include <asm/opal.h>
>> #include <linux/err.h>
>> +#include <asm/cputhreads.h>
>>
>> #define MAX_ATTR_LEN 32
>> #define MAX_LABEL_LEN 64
>> @@ -110,12 +111,54 @@ static ssize_t show_label(struct device
>> return sprintf(buf, "%s\n", sdata->label);
>> }
>>
>> +static int __init get_logical_cpu(unsigned int hwcpu)
>> +{
>> + int cpu;
>> +
>> + for_each_possible_cpu(cpu)
>> + if (get_hard_smp_processor_id(cpu) == hwcpu)
>> + return cpu;
>> +
>> + pr_err("%s: could not find a cpu with physical id 0x%x\n",
>> + __func__, hwcpu);
>
> I like moving this into a function, but I really dislike this error
> message. If the devicetree data is wrong/bad, the log and the console
> will be clogged with that message. And the user will not be able
> to do anything about it.
yes. It is not super useful anyway and it is redundant with the "phy"
prefix below. Do you want me to kill it in a v4 ?
Thanks,
C.
>
> Guenter
>
>> + return -ENOENT;
>> +}
>> +
>> static void __init make_sensor_label(struct device_node *np,
>> struct sensor_data *sdata, const char *label)
>> {
>> + u32 id;
>> size_t n;
>>
>> n = snprintf(sdata->label, sizeof(sdata->label), "%s", label);
>> +
>> + /*
>> + * Core temp pretty print
>> + */
>> + if (!of_property_read_u32(np, "ibm,pir", &id)) {
>> + int cpuid = get_logical_cpu(id);
>> +
>> + if (cpuid >= 0)
>> + /*
>> + * The digital thermal sensors are associated
>> + * with a core. Let's print out the range of
>> + * cpu ids corresponding to the hardware
>> + * threads of the core.
>> + */
>> + n += snprintf(sdata->label + n,
>> + sizeof(sdata->label) - n, " %d-%d",
>> + cpuid, cpuid + threads_per_core - 1);
>> + else
>> + n += snprintf(sdata->label + n,
>> + sizeof(sdata->label) - n, " phy%d", id);
>> + }
>> +
>> + /*
>> + * Membuffer pretty print
>> + */
>> + if (!of_property_read_u32(np, "ibm,chip-id", &id))
>> + n += snprintf(sdata->label + n, sizeof(sdata->label) - n,
>> + " %d", id & 0xffff);
>> }
>>
>> static int get_sensor_index_attr(const char *name, u32 *index,
>>
>>
>
_______________________________________________
lm-sensors mailing list
lm-sensors@xxxxxxxxxxxxxx
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
