Kernel thread scheduling

[ruben@mrbrklyn.com (Ruben Safir)]

On 04/11/2015 10:21 PM, Ruben Safir wrote:
> On 04/10/2015 09:09 AM, nick wrote:
>>
>>
>> On 2015-04-09 11:37 PM, Ruben Safir wrote:
>>> On 04/09/2015 10:52 PM, nick wrote:
>>>> Before asking questions again like this please look into either using lxr or ctags
>>>> to navigate the kernel tree for answers as can be faster then waiting for me or 
>>>> someone else to respond.
>>>
>>>
>>> well, I reading the text is ctags aren't much value there.
>>>
>> Ctags is useful for searching the code, which is why I am recommending it.
>> Nick
> 
> I have it built into gvim, but you can't use it from a textbook.  I'm
> finding it is not as useful as it could be for the kernel code.  There
> are stacks of tags to get around.  Another 2 days to learn to get around
> tags in vi is not in the agenda right now.  It is the tool I have so
> I'll have to live with it right now.
> 
> I also have a question that is not obvious from the code I'm looking at.
>  I'm not sure how these structs are attached together.  Or more
> specifically, I'm not sure how pulling the correct sched_entity gets one
> the coresponding task_entity
> 
> You have
> struct task_struct with a
> 	struct sched_entity
> 
> struct sched_enitities are nodes in the RB tree
> 	which are a "container" for "struct rb_node run_node".
> 
> So a look at sched_entity ... is in ../linux/sched.h
> 
> 1161 struct sched_entity {
> 1162    struct load_weight   load;    /* for load-balancing */
> 1163    struct rb_node    run_node;
> 1164    struct list_head  group_node;
> 1165    unsigned int      on_rq;
> 1166
> 1167    u64         exec_start;
> 1168    u64         sum_exec_runtime;
> 1169    u64         vruntime;
> 1170    u64         prev_sum_exec_runtime;
> 1171
> 1172    u64         nr_migrations;
> 1173
> 1174 #ifdef CONFIG_SCHEDSTATS
> 1175    struct sched_statistics statistics;
> 1176 #endif
> 1177
> 1178 #ifdef CONFIG_FAIR_GROUP_SCHED
> 1179    int         depth;
> 1180    struct sched_entity  *parent;
> 1181    /* rq on which this entity is (to be) queued: */
> 1182    struct cfs_rq     *cfs_rq;
> 1183    /* rq "owned" by this entity/group: */
> 1184    struct cfs_rq     *my_q;
> 1185 #endif
> 1186
> 1187 #ifdef CONFIG_SMP
> 1188    /* Per-entity load-tracking */
> 1189    struct sched_avg  avg;
> 1190 #endif
> 1191 };
> 
> I see no means of referencing a specific task from this struct that
> forms the node.  So when you pull the node with the smallest vruntime
> from the left most postion of the RB tree, by calling pick_next_task(),
> 
> 
> static struct sched_entity *__pick_next_entity(struct sched_entity *se)
> {
> 	struct rb_node *next = rb_next(&se->run_node);
> 
> 	if (!next)
> 		return NULL;
> 
> 	return rb_entry(next, struct sched_entity, run_node);
> }
> 
> 
> how do we know what task we are attached to?
> 
> Ruben
> 
> 

I'm still loss on how we know which taks_struct is being used but as a
side note, I found this also very puzzling

return rb_entry(next, struct sched_entity, run_node);
With help I ran it down to this:

http://lxr.free-electrons.com/source/include/linux/rbtree.h#L50

#define rb_entry(ptr, type, member) container_of(ptr, type, member)

which leads me to yet another macro

798 #define container_of(ptr, type, member) ({                      \
799         const typeof( ((type *)0)->member ) *__mptr = (ptr);    \
800         (type *)( (char *)__mptr - offsetof(type,member) );})


This is a use of macros I'd never seen before up close.  If anyone could
help me understand it, I'd appreciate it.

Ruben
> 
> 
>>> _______________________________________________
>>> Kernelnewbies mailing list
>>> Kernelnewbies at kernelnewbies.org
>>> http://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies
>>>
>>
>>
> 
> 
> _______________________________________________
> Kernelnewbies mailing list
> Kernelnewbies at kernelnewbies.org
> http://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies
> 
>