Re: [ckrm-tech] [RFC][PATCH 6/8] RSS controller shares allocation

[Balbir Singh <balbir@in.ibm.com>]

Pavel Emelianov wrote:
> Balbir Singh wrote:
>> Support shares assignment and propagation.
>>
>> Signed-off-by: Balbir Singh <balbir@in.ibm.com>
>> ---
>>
>>  kernel/res_group/memctlr.c |   59 ++++++++++++++++++++++++++++++++++++++++++++-
>>  1 file changed, 58 insertions(+), 1 deletion(-)
> 
> [snip]
> 
>> +static void recalc_and_propagate(struct memctlr *res, struct memctlr *parres)
>> +{
>> +	struct resource_group *child = NULL;
>> +	int child_divisor;
>> +	u64 numerator;
>> +	struct memctlr *child_res;
>> +
>> +	if (parres) {
>> +		if (res->shares.max_shares == SHARE_DONT_CARE ||
>> +			parres->shares.max_shares == SHARE_DONT_CARE)
>> +			return;
>> +
>> +		child_divisor = parres->shares.child_shares_divisor;
>> +		if (child_divisor == 0)
>> +			return;
>> +
>> +		numerator = (u64)(parres->shares.unused_min_shares *
>> +				res->shares.max_shares);
>> +		do_div(numerator, child_divisor);
>> +		numerator = (u64)(parres->nr_pages * numerator);
>> +		do_div(numerator, SHARE_DEFAULT_DIVISOR);
>> +		res->nr_pages = numerator;
>> +	}
>> +
>> +	for_each_child(child, res->rgroup) {
>> +		child_res = get_memctlr(child);
>> +		BUG_ON(!child_res);
>> +		recalc_and_propagate(child_res, res);
> 
> Recursion? Won't it eat all the stack in case of a deep tree?

The depth of the hierarchy can be controlled. Recursion is needed
to do a DFS walk

> 
>> +	}
>> +
>> +}
>> +
>> +static void memctlr_shares_changed(struct res_shares *shares)
>> +{
>> +	struct memctlr *res, *parres;
>> +
>> +	res = get_memctlr_from_shares(shares);
>> +	if (!res)
>> +		return;
>> +
>> +	if (is_res_group_root(res->rgroup))
>> +		parres = NULL;
>> +	else
>> +		parres = get_memctlr((struct container *)res->rgroup->parent);
>> +
>> +	recalc_and_propagate(res, parres);
>> +}
>> +
>>  struct res_controller memctlr_rg = {
>>  	.name = res_ctlr_name,
>>  	.ctlr_id = NO_RES_ID,
>>  	.alloc_shares_struct = memctlr_alloc_instance,
>>  	.free_shares_struct = memctlr_free_instance,
>>  	.move_task = memctlr_move_task,
>> -	.shares_changed = NULL,
>> +	.shares_changed = memctlr_shares_changed,
> 
> I didn't find where in this patches this callback is called.

It's a part of the resource groups infrastructure. It's been ported
on top of Paul Menage's containers patches. The code can be easily
adapted to work directly with containers instead of resource groups
if required.



-- 

	Balbir Singh,
	Linux Technology Center,
	IBM Software Labs
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/