Re: [PATCH 1/3 v6] cpufreq: Add debugfs directory for cpufreq

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 07/24/2013 04:51 PM, Viresh Kumar wrote:
> On 24 July 2013 13:13, Chanwoo Choi <cw00.choi@xxxxxxxxxxx> wrote:
>> On 07/24/2013 02:05 PM, Viresh Kumar wrote:
>>> On 24 July 2013 06:55, Chanwoo Choi <cw00.choi@xxxxxxxxxxx> wrote:
>>>> On 07/22/2013 07:11 PM, Viresh Kumar wrote:
>>>>> On 18 July 2013 16:47, Chanwoo Choi <cw00.choi@xxxxxxxxxxx> wrote:
>>>
>>>>>> +static void cpufreq_remove_debugfs_dir(struct cpufreq_policy *policy,
>>>>>> +                                      unsigned int cpu)
>>>>>> +{
>>>>>> +       unsigned int idx = cpumask_weight(policy->cpus) > 1 ? cpu : 0;
>>>>>> +
>>>>>> +       if (!policy->cpu_debugfs[idx])
>>>>>> +               return;
>>>>>> +
>>>>>> +       debugfs_remove_recursive(policy->cpu_debugfs[idx]);
>>>>>
>>>>> Whey do we need recursive here? And what exactly does recursive will
>>>>> do?
>>>>>
>>>>
>>>> If cpu is last user of policy, __cpufreq_remove_dev() have to remove debugfs directory
>>>> and child file/directory of root debugfs directory. So, I used debugfs_remove_recursive() function.
>>>
>>> You are calling this routine even when we aren't at the last cpu of a policy.
>>> And so, eventually you are calling this routine for a link you have created.
>>
>> I'll call proper debugfs_remove*() function according to type of debugfs pointer.
>> - if cpu is last user of policy, call debugfs_remove_recursive()
>> - else, call debugfs_remove().
>>
>>>
>>> Have you actually tested your code? What kind of platform? What is cpu
>>> topology ?? And what exactly you tested..
>>
>> I tested quad-core EXYNOS4412 SoC based on Cortex-A9 with Tizen platform.
>> It is opereated on this environment but as you commnet, this test and environment
>> isn't enough to verify this patchset.
>> - Testcase1 : Change cpufreq governor on runtime
>> - Testcase2 : Turn on/off CPU state on runtme
>>
>>>
>>> We are already on v6 and this patch still looks like the v1.. It still has lots
>>> of basic mistakes, which I don't expect so later in the series..
>>>
>>> Its very difficult for me to review the same patchset again and again.. So,
>>> normally people might not review it well after v3-v4 and just trust the sender..
>>> But I am nowhere close to getting that.. And so discouraged to review it..
>>>
>>
>> I'm so sorry about this and thanks for previous your review sincerely.
>>
>>> Please review/test it well on multiple kind of systems if possible. Test on
>>> your intel laptop and see if it has multiple policy structures with
>>> multiple cpus
>>> in it.. cpuX/cpufreq/related_cpus gives you all cpus that share policy
>>> structure.
>>
>> As you comment, I'll modify/test this patchset on various system with enough testcase
>> and resend this patchset after a thorough review.
>>
>>
>>>
>>>>>> +}
>>>>>> +
>>>>>
>>>>> same problem here too.
>>>>>> +static void cpufreq_move_debugfs_dir(struct cpufreq_policy *policy,
>>>>>> +                                    unsigned int new_cpu)
>>>>>> +{
>>>>>> +       struct dentry *old_entry, *new_entry;
>>>>>> +       char new_dir_name[CPUFREQ_NAME_LEN];
>>>>>> +       unsigned int j, old_cpu = policy->cpu;
>>>>>> +
>>>>>> +       if (!policy->cpu_debugfs[new_cpu])
>>>>>> +               return;
>>>>>> +
>>>>>> +       /*
>>>>>> +        * Remove symbolic link of debugfs directory except for debugfs
>>>>>> +        * directory of old_cpu.
>>>>>> +        */
>>>>>> +       for_each_present_cpu(j) {
>>>>>> +               if (old_cpu == j)
>>>>>> +                       continue;
>>>>>> +
>>>>>> +               debugfs_remove(policy->cpu_debugfs[j]);
>>>>>
>>>>> Why you need this? We aren't removing the earlier dentry at all here.
>>>
>>> haven't answered this.
>>
>> The debugfs entry of 'old_cpu' include child debugfs file(e.g., load_table)
>> If cpu is last user of policy and core call __cpufre_remove_dev() to remove last cpu,
>> core call cpufreq_move_debugfs_dir(). I have to move the data of debugfs directory/file and
>> child data for 'old_cpu' to debugfs directory for 'new_cpu'.
>>
>> If I remove earlier dentry of 'old_cpu', I can't get the child debugfs dir/file.
>> So I didn't remove the earlier dentry of 'old_cpu'.
>>
>>>
>>>>>> +       if (!new_entry) {
>>>>>> +               pr_err("changing debugfs directory name failed\n");
>>>>>> +               goto err_rename;
>>>>>> +       }
>>>>>> +
>>>>>> +       policy->cpu_debugfs[new_cpu] = new_entry;
>>>>>> +       policy->cpu_debugfs[old_cpu] = NULL;
>>>>>> +
>>>>>> +       /* Create again symbolic link of debugfs directory */
>>>>>> +       for_each_present_cpu(j) {
>>>>>
>>>>> present_cpu?? We discussed this before.. You will break multi cluster
>>>>> systems.
>>>>
>>>> My mistake. I'll use for_each_cpu() macro instead of for_each_present_cpu().
>>>
>>> Go through earlier comments about this.. you are still wrong.. You need to
>>> run over cpus that are in this policy.. i.e. policy->cpus.
>>>
>>
>> OK.
>>
>>>>>> +               if (new_cpu == j)
>>>>>> +                       continue;
>>>>>> +
>>>
>>>>>> @@ -1894,6 +2065,8 @@ int cpufreq_register_driver(struct cpufreq_driver *driver_data)
>>>>>>         cpufreq_driver = driver_data;
>>>>>>         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
>>>>>>
>>>>>> +       cpufreq_create_debugfs();
>>>>>
>>>>> Why you moved this to register_driver? It was fine at cpufreq_core_init()
>>>>
>>>> If we moved this to cpufreq_core_int(), I have to create cpufreq_core_exit().
>>>> Do you agree about creating cpufreq_core_exit()(?
>>>
>>> No you don't need that routine. Or in other words there isn't any exit
>>> for cpufreq core and so this directory must not be removed.
>>>
>>
>> I understood on your previous comment as You said that I had to remove 'cpufreq' debugfs directory
>> when cpufreq isn't used.
>>
>> If the core execute cpufreq_create_debugfs() in cpufreq_core_init(),
>> don't I need to remove 'cpufreq' debugfs directory without cpufreq_core_exit()?
> 
> I copied following from your patch sent on 5th july.. It didn't had any version
> number and so is difficult to distinguish..
> 
>> @@ -1976,6 +2029,10 @@ static int __init cpufreq_core_init(void)
>>         BUG_ON(!cpufreq_global_kobject);
>>         register_syscore_ops(&cpufreq_syscore_ops);
>>
>> +       cpufreq_debugfs = debugfs_create_dir("cpufreq", NULL);
>> +       if (!cpufreq_debugfs)
>> +               pr_debug("creating debugfs root failed\n");
> 
> So, you just added this directory once.. So you must not
> remove it.
> 
> 
> Where did I say you remove this directory..
> To be clear, don't remove cpufreq debugfs directory at all. Play
> only with cpu directories inside this debugfs directory.
> 

You're right. I'm sorry and I misunderstood.

Thanks,
Chanwoo Choi
--
To unsubscribe from this list: send the line "unsubscribe cpufreq" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html




[Index of Archives]     [Linux Kernel Devel]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite Forum]     [Linux SCSI]

  Powered by Linux