On 14/08/13 13:53, Rob Herring wrote:
> On 08/14/2013 05:01 AM, Sudeep KarkadaNagesha wrote:
>> On 13/08/13 22:07, Benjamin Herrenschmidt wrote:
>>> On Tue, 2013-08-13 at 19:29 +0100, Sudeep KarkadaNagesha wrote:
>>>> I don't understand completely the use of ibm,ppc-interrupt-server#s and
>>>> its implications on generic of_get_cpu_node implementation.
>>>> I see the PPC specific definition of of_get_cpu_node uses thread id only
>>>> in 2 instances. Based on that, I have tried to move all the other
>>>> instances to use generic definition.
>>>>
>>>> Let me know if the idea is correct.
>>>
>>> No. The device-tree historically only represents cores, not HW threads, so
>>> it makes sense to retrieve also the thread number corresponding to the CPU.
>>>
>> Ok
>>
>>> However, the mechanism to represent HW threads in the device-tree is currently
>>> somewhat platform specific (the ibm,ppc-interrupt-server#s).
>> I see most of the callers pass NULL to thread id argument except 2
>> instances in entire tree. If that's the case why can't we move to use
>> generic of_get_cpu_node in most of those cases and have PPC specific
>> implementation for the ones using thread id.
>>
>>>
>>> So what you could do for now is:
>>>
>>> - Have a generic version that always returns 0 as the thread, which is weak
>> I would prefer to move to generic of_get_cpu_node where ever possible
>> and rename the function that takes thread id rather than making generic
>> one weak.
>>
>>>
>>> - powerpc keeps its own implementation
>> How about only in cases where it needs thread_id.
>>
>>>
>>> - Start a discussion on the bindings (if not already there) to define threads
>>> in a better way at which point the generic function can be updated.
>>>
>> I am not sure if we need to define any new bindings. Excerpts from ePAPR
>> (v1.1):
>> "3.7.1 General Properties of CPU nodes
>> The value of "reg" is a <prop-encoded-array> that defines a unique
>> CPU/thread id for the CPU/threads represented by the CPU node.
>> If a CPU supports more than one thread (i.e. multiple streams of
>> execution) the reg property is an array with 1 element per thread. The
>> #address-cells on the /cpus node specifies how many cells each element
>> of the array takes. Software can determine the number of threads by
>> dividing the size of reg by the parent node's #address-cells."
>>
>> And this is exactly in agreement to what's implemented in the generic
>> of_get_cpu_node:
>>
>> for_each_child_of_node(cpus, cpun) {
>> if (of_node_cmp(cpun->type, "cpu"))
>> continue;
>> cell = of_get_property(cpun, "reg", &prop_len);
>> if (!cell) {
>> pr_warn("%s: missing reg property\n", cpun->full_name);
>> continue;
>> }
>> prop_len /= sizeof(*cell);
>> while (prop_len) {
>> hwid = of_read_number(cell, ac);
>> prop_len -= ac;
>> if (arch_match_cpu_phys_id(cpu, hwid))
>> return cpun;
>> }
>> }
>
> How about something like this:
>
> for_each_child_of_node(cpus, cpun) {
> if (of_node_cmp(cpun->type, "cpu"))
> continue;
>
> if (arch_of_get_cpu_node(cpun, thread))
> return cpun;
>
> cell = of_get_property(cpun, "reg", &prop_len);
> if (!cell) {
> pr_warn("%s: missing reg property\n", cpun->full_name);
> continue;
> }
> prop_len /= sizeof(*cell);
> while (prop_len) {
> hwid = of_read_number(cell, ac);
> prop_len -= ac;
> if (arch_match_cpu_phys_id(cpu, hwid))
> return cpun;
> }
> }
>
> For PPC:
>
> arch_of_get_cpu_node()
> {
> const u32 *intserv;
> unsigned int plen, t;
>
> /* Check for ibm,ppc-interrupt-server#s. */
> intserv = of_get_property(np, "ibm,ppc-interrupt-server#s",
> &plen);
> if (!intserv)
> return false;
>
> hardid = get_hard_smp_processor_id(cpu);
>
> plen /= sizeof(u32);
> for (t = 0; t < plen; t++) {
> if (hardid == intserv[t]) {
> if (thread)
> *thread = t;
> return true;
> }
> }
> return false;
> }
>
Sorry responded to earlier mail before seeing this. This approach looks
good, but we still need to have thread id as argument which should be fine.
But as per my understanding on how logical cpu<->hard proccessor id is
setup, the thread_id is implicit in the logical cpu id making it
unnecessary to depend on DT each time.
Regards,
Sudeep
>>
>> Yes this doesn't cover the historical "ibm,ppc-interrupt-server#s", for
>> which we can have PPC specific wrapper above the generic one i.e. get
>> the cpu node and then parse for thread id under custom property.
>>
>> Let me know your thoughts.
>>
>> Regards,
>> Sudeep
>>
>>
>>
>
>
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
