Change Details

We currently read the pdwakeups counter to determine how many inactive queue scans were performed by the page daemon. This is used to determine whether to start a background laundering. However, this was done before the page daemon had a mechanism to wake up the laundry thread, notifying it of a scan. It's a bit weird to read a per-CPU counter for this purpose, and I'd like to propose a change which makes it possible for the page daemon to scan the inactive queue multiple times without sleeping, which would make pdwakeups unreliable. So let's use a dedicated counter.

We currently read the pdwakeups counter to determine how many inactive queue scans were performed by the page daemon. This is used to determine whether to start a background laundering. However, this was done before the page daemon had a mechanism to wake up the laundry thread, notifying it that it had just performed a scan. It's a bit weird to read a per-CPU counter for this purpose, and I'd like to propose a change which makes it possible for the page daemon to scan the inactive queue multiple times without sleeping, which would make pdwakeups unreliable. So let's use a dedicated counter, synchronized by the PQ_LAUNDRY queue lock.

We currently read the pdwakeups counter to determine how many inactive queue scans were performed by the page daemon. This is used to determine whether to start a background laundering. However, this was done before the page daemon had a mechanism to wake up the laundry thread, notifying it ofit that it had just performed a scan. It's a bit weird to read a per-CPU counter for this purpose, and I'd like to propose a change which makes it possible for the page daemon to scan the inactive queue multiple times without sleeping, which would make pdwakeups unreliable. So let's use a dedicated counter, synchronized by the PQ_LAUNDRY queue lock.