LWN: Comments on "Write-protect for userfaultfd()"

Write-protect for userfaultfd()

robert_s — Thu, 16 May 2019 23:33:00 +0000

*Ahem* Software Transactional Memory

Write-protect for userfaultfd()

nix — Mon, 06 May 2019 22:55:09 +0000

Quite. Getting this wrong leads to hilariously terrible disasters: https://web.archive.org/web/20180903022426/http://www.os2...

PTE bits

corbet — Fri, 03 May 2019 11:01:23 +0000

The PTE format is defined by the hardware (on many architectures, at least); it's not something that can be readily extended.

Write-protect for userfaultfd()

matthias — Fri, 03 May 2019 11:00:46 +0000

Because the size (and structure) of page table entries is defined by the hardware. After all, the MMU has to be able to decode the entries when converting virtual to physical addresses. There are only a few bits that can be used by the operating system and that are ignored by the hardware.

Write-protect for userfaultfd()

Freeaqingme — Fri, 03 May 2019 07:14:08 +0000

Why does it matter that the last available bit of the page table entry would be used? It's not something that needs to be backwards compatible, right? Besides a (very tiny?) bit of performance loss, what would be the argument against increasing it with a byte when the next feature comes around?

Write-protect for userfaultfd()

unixbhaskar — Thu, 02 May 2019 22:56:06 +0000

:) :) Andrew!! " "Always" was the inevitable answer from memory-management subsystem maintainer Andrew Morton." ...aha!