[ltt-dev] [PATCH 9/9] change strategy for AR_RANDOM

[Lai Jiangshan]

On 11/15/2011 07:35 PM, Mathieu Desnoyers wrote:
> * Lai Jiangshan (laijs at cn.fujitsu.com) wrote:
>> Signed-off-by: Lai Jiangshan <laijs at cn.fujitsu.com>
>> ---
>>  tests/test_urcu_hash.c |   34 ++++++++++++++++++++++++++++++++--
>>  1 files changed, 32 insertions(+), 2 deletions(-)
>>
>> diff --git a/tests/test_urcu_hash.c b/tests/test_urcu_hash.c
>> index 964a4e1..f07f0d4 100644
>> --- a/tests/test_urcu_hash.c
>> +++ b/tests/test_urcu_hash.c
>> @@ -540,6 +540,37 @@ void free_node_cb(struct rcu_head *head)
>>  	free(node);
>>  }
>>  
>> +static inline
>> +int thr_writer_add(void)
>> +{
>> +	static __thread int addremove_thread;
>> +	unsigned addremove_random;
>> +
>> +	if (add_only)
>> +		return 1;
>> +
>> +	if (addremove == AR_ADD)
>> +		return 1;
>> +
>> +	if (addremove == AR_REMOVE)
>> +		return 0;
>> +
>> +	addremove_random = test_rand_get_bits(2);
>> +	if (addremove_random < 2)
>> +		return addremove_random;
>> +
>> +	while (caa_unlikely(!addremove_thread))
>> +		addremove_thread = test_rand_get_bits(16) - (1UL << 15);
>> +
>> +	if (addremove_thread > 0) {
>> +		addremove_thread--;
>> +		return 1;
>> +	} else {
>> +		addremove_thread++;
>> +		return 0;
>> +	}
> 
> Can you explain what this part of the change does ? 

It only change the behavior when AR_RANDOM.

I want to test this case:
add     a lot of nodes to the hash table but delete seldom in period in a thread.
delete                                       add


> I am slightly
> uncomfortable about a while () loop in there too that depends on random
> results.
> 


I can change to

if (caa_unlikely(!addremove_thread)) {
	addremove_thread = test_rand_get_bits(16) - (1UL << 15);
	if (caa_unlikely(!addremove_thread))
		return test_rand_get_bits(1);
}

Thanks,
Lai