Re: [PATCH V2 3/3] sched/numa: Reset the accessing PID information periodically

[Raghavendra K T]

On 2/3/2023 5:05 PM, Peter Zijlstra wrote:
> On Wed, Feb 01, 2023 at 01:32:22PM +0530, Raghavendra K T wrote:
>
> > 2) Maintain duplicate list of accessing PIDs to keep track of history of access. and switch/reset. use OR operation during iteration
> >
> >   Two lists of PIDs maintained. At regular interval old list is reset and we make current list as old list
> > At any point of time tracking of PIDs accessing VMA is determined by ORing list1 and list2
> >
> > accessing_pids_list1 <-  current list
> > accessing_pids_list2 <-  old list
>
> ( I'm not sure why you think this part of the email doesn't need to be
>    nicely wrapped at 76 chars.. )

Sorry.. copy pasted from my "idea" notes then word wrap fooled me..

> This seems simple enough to me and can be trivially extended to N if
> needed.
>  > The typical implementation would looks something like:
>
> 	unsigned long pids[N];
> 	unsigned int pid_idx;
>
> set:
> 	unsigned long *pids = numab->pids + pid_idx;
> 	if (!__test_bit(bit, pids))
> 		__set_bit(bit, pids);
>
> test:
> 	unsigned long pids = 0;
> 	for (int i = 0; i < N; i++)
> 		pids |= numab->pids[i];
> 	return __test_bit(bit, &pids);
>
> rotate:
> 	idx = READ_ONCE(numab->pid_idx);
> 	WRITE_ONCE(numab->pid_idx, (idx + 1) % N);
> 	numab->pids[idx] = 0;
>
> Note the actual rotate can be simplified to ^1 for N:=2.

Thanks good idea. This will be very helpful when we want to
differentiate accessing PIDs in more granular way. Perhaps we can go
with N=2 and stick to below simplification of your code above?

something like:

unsigned long pids[2]

// Assume pids[1] has latest detail always
set:
if (!__test_bit(bit, pids[1])
	__set_bit(bit, pids[1])

test:
unsigned long pids = pids[0] | pids[1];
return __test_bit(bit, &pids);

rotate:
pids[0] = pids[1];
pids[1] = 0;