Re: [PATCH v1 2/2] mm: mTHP stats for pagecache folio allocations

From: Ryan Roberts
Date: Sat Jul 13 2024 - 07:00:38 EST


[...]

>> +static int thpsize_create(int order, struct kobject *parent)
>>   {
>>       unsigned long size = (PAGE_SIZE << order) / SZ_1K;
>> +    struct thpsize_child *stats;
>>       struct thpsize *thpsize;
>>       int ret;
>>   +    /*
>> +     * Each child object (currently only "stats" directory) holds a
>> +     * reference to the top-level thpsize object, so we can drop our ref to
>> +     * the top-level once stats is setup. Then we just need to drop a
>> +     * reference on any children to clean everything up. We can't just use
>> +     * the attr group name for the stats subdirectory because there may be
>> +     * multiple attribute groups to populate inside stats and overlaying
>> +     * using the name property isn't supported in that way; each attr group
>> +     * name, if provided, must be unique in the parent directory.
>> +     */
>> +
>>       thpsize = kzalloc(sizeof(*thpsize), GFP_KERNEL);
>> -    if (!thpsize)
>> -        return ERR_PTR(-ENOMEM);
>> +    if (!thpsize) {
>> +        ret = -ENOMEM;
>> +        goto err;
>> +    }
>> +    thpsize->order = order;
>>         ret = kobject_init_and_add(&thpsize->kobj, &thpsize_ktype, parent,
>>                      "hugepages-%lukB", size);
>>       if (ret) {
>>           kfree(thpsize);
>> -        return ERR_PTR(ret);
>> +        goto err;
>>       }
>>   -    ret = sysfs_create_group(&thpsize->kobj, &thpsize_attr_group);
>> -    if (ret) {
>> +    stats = kzalloc(sizeof(*stats), GFP_KERNEL);
>> +    if (!stats) {
>>           kobject_put(&thpsize->kobj);
>> -        return ERR_PTR(ret);
>> +        ret = -ENOMEM;
>> +        goto err;
>>       }
>>   -    ret = sysfs_create_group(&thpsize->kobj, &stats_attr_group);
>> +    ret = kobject_init_and_add(&stats->kobj, &thpsize_child_ktype,
>> +                   &thpsize->kobj, "stats");
>> +    kobject_put(&thpsize->kobj);
>>       if (ret) {
>> -        kobject_put(&thpsize->kobj);
>> -        return ERR_PTR(ret);
>> +        kfree(stats);
>> +        goto err;
>>       }
>>   -    thpsize->order = order;
>> -    return thpsize;
>> +    if (BIT(order) & THP_ORDERS_ALL_ANON) {
>> +        ret = sysfs_create_group(&thpsize->kobj, &thpsize_attr_group);
>> +        if (ret)
>> +            goto err_put;
>> +
>> +        ret = sysfs_create_group(&stats->kobj, &stats_attr_group);
>> +        if (ret)
>> +            goto err_put;
>> +    }
>> +
>> +    if (BIT(order) & PAGECACHE_LARGE_ORDERS) {
>> +        ret = sysfs_create_group(&stats->kobj, &file_stats_attr_group);
>> +        if (ret)
>> +            goto err_put;
>> +    }
>> +
>> +    list_add(&stats->node, &thpsize_child_list);
>> +    return 0;
>> +err_put:
>
> IIUC, I think you should call 'sysfs_remove_group' to remove the group before
> putting the kobject.

Are you sure about that? As I understood it, sysfs_create_group() was
conceptually modifying the state of the kobj, so when the kobj gets destroyed,
all its state is tidied up. __kobject_del() (called on the last kobject_put())
calls sysfs_remove_groups() and tidies up the sysfs state as far as I can see?

>
>> +    kobject_put(&stats->kobj);
>> +err:
>> +    return ret;
>>   }