On Wed, Sep 11, 2024 at 09:32:29PM -0400, Liam R. Howlett wrote: > * Helge Deller <deller@xxxxxx> [240911 20:51]: > > On 9/12/24 01:05, Liam R. Howlett wrote: > > > If someone is using MAP_STACK to avoid having a huge page, they will > > > also get a mapping that grows - which is different than what happens > > > today. > > > > > > I'm not saying that's right, but someone could be abusing the existing > > > flag and this will change the behaviour. > > > > Wouldn't a plain mmap() followed by madvise(MADV_NOHUGEPAGE) do exactly that? > > Why abusing MAP_STACK for that? > > I can think of two answers: > 1. An error that has worked without issues so far > 2. One less system call > > I'm not saying this really is a blocker, but the change is not without > risk as it does change behaviour the user could see. > > Interestingly enough, the man page is incorrect as it is written because > the flag is not strictly a no-op; it ensures no huge pages. So the > feature of applying VM_NOHUGEPAGE with the use of MAP_STACK is not > documented today. It's a recent change and I don't think it's something we necessarily want to document. It was c4608d1bf7c6 which was December 2023.
