Re: [PATCH -next v4] staging: fwserial: Switch to kfree_rcu() API

[shangxiaojing <shangxiaojing@huawei.com>]

On 2022/9/19 17:11, Greg KH wrote:
> On Mon, Sep 19, 2022 at 05:10:56PM +0800, Shang XiaoJing wrote:
>> Instead of invoking a synchronize_rcu() to free a pointer after a grace
>> period, we can directly make use of a new API that does the same but in
>> a more efficient way.
>>
>> Signed-off-by: Shang XiaoJing <shangxiaojing@huawei.com>
>> ---
>> Changelog:
>> v3: the first version of the PATCH
>> v1: v3 resent as v1
>> v2: use kfree_rcu() instead of kvfree_rcu() for clarity
>> v4: resend v2 as v4 to avoid versioning confusion
>> ---
>>   drivers/staging/fwserial/fwserial.c | 3 +--
>>   1 file changed, 1 insertion(+), 2 deletions(-)
>>
>> diff --git a/drivers/staging/fwserial/fwserial.c b/drivers/staging/fwserial/fwserial.c
>> index 81b06d88ed0d..8d2b4ed1f39e 100644
>> --- a/drivers/staging/fwserial/fwserial.c
>> +++ b/drivers/staging/fwserial/fwserial.c
>> @@ -2117,8 +2117,7 @@ static void fwserial_remove_peer(struct fwtty_peer *peer)
>>   	if (port)
>>   		fwserial_release_port(port, true);
>>   
>> -	synchronize_rcu();
>> -	kfree(peer);
>> +	kfree_rcu(peer);

The kfree_rcu(peer) should be kfree_rcu(peer, rcu), due to the rcu_head 
member named rcu in fwtty_peer.

This will be fixed in v5, sorry.

> What is "more efficient" about this change?

kfree_rcu() will call kvfree_call_rcu, due to the note in kernel/rcu/tree.c:

/Each kvfree_call_rcu() request is added to a batch. The batch will be 
drained every KFREE_DRAIN_JIFFIES number of jiffies. All the objects in 
the batch will be free'd in workqueue context. This allows us to: batch 
requests together to reduce the number of grace periods during heavy 
kfree_rcu()/kvfree_rcu() load. ///

which is the reason I use "more efficient".
>
> And do you have the hardware for this device to test with?  It would be
> good to finally get this out of staging, or to just remove it entirely
> if no one uses it.
Actually, not. I just found the code here may be optimized.
>
> thanks,
>
> greg k-h

Thanks,

Shang XiaoJing