Hi, On 12/7/2022 10:58 AM, Alexei Starovoitov wrote: > On Tue, Dec 6, 2022 at 6:20 PM Hou Tao <houtao@xxxxxxxxxxxxxxx> wrote: >> Hi, >> >> On 12/7/2022 9:52 AM, Yonghong Song wrote: >>> >>> On 12/5/22 8:29 PM, Hou Tao wrote: >>>> From: Hou Tao <houtao1@xxxxxxxxxx> >>>> >>>> When there are batched freeing operations on a specific CPU, part of >>>> the freed elements ((high_watermark - lower_watermark) / 2 + 1) will >>>> be moved to waiting_for_gp list and the remaining part will be left in >>>> free_by_rcu list and waits for the expiration of RCU-tasks-trace grace >>>> period and the next invocation of free_bulk(). >>> The change below LGTM. However, the above description seems not precise. >>> IIUC, free_by_rcu list => waiting_for_gp is controlled by whether >>> call_rcu_in_progress is true or not. If it is true, free_by_rcu list >>> will remain intact and not moving into waiting_for_gp list. >>> So it is not 'the remaining part will be left in free_by_rcu'. >>> It is all elements in free_by_rcu to waiting_for_gp or none. >> Thanks for the review and the suggestions. I tried to say that moving from >> free_by_rcu to waiting_for_gp is slow, and there can be many free elements being >> stacked on free_by_rcu list. So how about the following rephrasing or do you >> still prefer "It is all elements in free_by_rcu to waiting_for_gp or none." ? >> >> When there are batched freeing operations on a specific CPU, part of the freed >> elements ((high_watermark - lower_watermark) / 2 + 1) will be moved to >> waiting_for_gp list and the remaining part will be left in free_by_rcu list. > I agree with Yonghong. > The above sentence is not just not precise. > 'elements moved to waiting_for_gp list' part is not correct. > The elements never moved into it directly. > Only via free_by_rcu list. Yes. > > All or none also matters. I am still confused about the "All or none". Does it mean all elements in free_by_list will be moved into waiting_for_gp or none will be moved if call_rcu_in_progress is true, right ? So How about the following rephrasing ? When there are batched freeing operations on a specific CPU, part of the freed elements ((high_watermark - lower_watermark) / 2 + 1) will be indirectly moved into waiting_for_gp list through free_by_rcu list. After call_rcu_in_progress becomes false again, the remaining elements in free_by_rcu list will be moved to waiting_for_gp list by the next invocation of free_bulk(). However if the expiration of RCU tasks trace grace period is relatively slow, none element in free_by_rcu list will be moved. So instead of invoking __alloc_percpu_gfp() or kmalloc_node() to allocate a new object, in alloc_bulk() just check whether or not there is freed element in free_by_rcu list and reuse it if available. > .
