Messages in this thread |  | | Date | Fri, 16 Dec 2022 05:16:45 +0300 | From | kirill.shutemov@linux ... | Subject | Re: [GIT PULL] x86/mm for 6.2 |
| |
On Thu, Dec 15, 2022 at 09:17:11AM -0800, Linus Torvalds wrote: > - make thread creation lock it (and unlock it on creating a new mm at > execve, but not fork). > > And yes, I would actually suggest that _any_ thread creation locks it, > so that you never *EVER* have any issues with "oh, now I need to > synchronize with other threads". A process can set its LAM state at > startup, not in the middle of running!
By thread creation, you mean CLONE_VM, right?
Do we care to avoid locking for CLONE_VFORK case? Seems excessive.
> Note that io_uring will automatically do that locking, because it does > that thread creation (create_io_thread()). So io_uring threads won't > even be a special case.
I've missed that io_uring does not use kthread_use_mm() anymore. But what about other kthread_use_mm() users?
I did not find concrete cases where missing LAM setup would breaks things, but I cannot prove opposite too.
kthread_use_mm() usage is suspicious in amdkfd, but I donno.
io_uring was a clear before it moved away from kthread_use_mm().
-- Kiryl Shutsemau / Kirill A. Shutemov
|  |