LWN: Comments on "The ongoing search for mmap_lock scalability"

Typo reports

corbet — Mon, 09 May 2022 12:58:56 +0000

Fixed.

Next time, though, please send typo reports via email so that thousands of other LWN readers don't have to read through them.

The ongoing search for mmap_lock scalability

bgoglin — Mon, 09 May 2022 09:12:54 +0000

Typo: "It"->"It's" in "It about time to start ..." ?

Technically correction for Android app starting

lengfeld — Mon, 09 May 2022 09:10:30 +0000

> When Android runs an app, it does so by spawning a new thread which typically starts by mapping a number of VMAs.

When Android runs an app, it forks a new process from the zygote process. zygote contains an already stared Java VM including all the framework java libraries. Since fork() uses Copy-on-Write the parent and child process shared a lot of the memory pages. When the child starts to write to some of these shared pages, page faults occur, because the pages are mapped as read-only.

The ongoing search for mmap_lock scalability

willy — Sat, 07 May 2022 22:24:15 +0000

By the time we call the driver, we already allocated all the way down to the PMD level, so that's all fine.

In my plan, drivers that want to opt into RCU page fault handling get to implement map_pages. Although the other MM developers have asked that we not do full RCU page fault handling until it's proven necessary.

The ongoing search for mmap_lock scalability

luto — Sat, 07 May 2022 19:59:11 +0000

On x86 (and probably most or even all other architectures), the total memory that a page fault might need to allocate for page table filling purposes is bounded and quite small. It would seem entirely reasonable and not necessarily even particularly difficult for do_user_addr_fault to grab its five pages in advance and avoid GFP_KERNEL allocations. The only real issue is that the control flow is backwards - the actual peg table setup in called from driver code.

We could perhaps have a per-task lookaside list for these pages, though, or a per-task “don’t-use-GFP_KERNEL” flag.