On Tue, Jul 23, 2024 at 11:18 AM Christoph Hellwig <hch@xxxxxxxxxxxxx> wrote: > > On Mon, Jul 22, 2024 at 11:01:23AM +0200, Michal Hocko wrote: > > __GFP_$FOO are usually low level. GFP_$FOO are high level and they > > combine several subflags to have a specific meaning. So this would need > > to be GFP_NOFAIL. Btw. the same applies to __GFP_NORETRY and > > __GFP_RETRY_MAYFAIL. > > True. But I think adding GFP_NOFAIL and slowly upping the enforcement > that no one is using __GFP_NOFAIL directly will get us a similar effect > to my *_nofail proposal. It will require manual or scripted checking > instead of relying on the compiler, but it's much better than what we > have right now. I agree. My proposal was actually to enforce blocking in GFP_NOFAIL while the post code was an ugly hack at the lower level with __GFP_NOFAIL.
