On Fri, Feb 05, 2021 at 03:23:00PM -0800, ira.weiny@xxxxxxxxx wrote: > From: Ira Weiny <ira.weiny@xxxxxxxxx> > > There are many places where kmap/<operation>/kunmap patterns occur. We lift > these various patterns to core common functions and use them in the btrfs file > system. At the same time we convert those core functions to use > kmap_local_page() which is more efficient in those calls. > > I think this is best accepted through Andrew's tree as it has the mem*_page > functions in it. But I'd like to get an ack from David or one of the other > btrfs maintainers before the btrfs patches go through. I'd rather take the non-mm patches through my tree so it gets tested the same way as other btrfs changes, straightforward cleanups or not. This brings the question how to do that as the first patch should go through the MM tree. One option is to posptpone the actual cleanups after the 1st patch is merged but this could take a long delay. I'd suggest to take the 1st patch within MM tree in the upcoming merge window and then I can prepare a separate pull with just the cleanups. Removing an inter-tree patch dependency was a sufficient reason for Linus in the past for such pull requests. > There are a lot more kmap->kmap_local_page() conversions but kmap_local_page() > requires some care with the unmapping order and so I'm still reviewing those > changes because btrfs uses a lot of loops for it's kmaps. It sounds to me that converting the kmaps will take some time anyway so exporting the helpers first and then converting the subsystems might be a good option. In case you'd like to get rid of the simple cases in btrfs code now we can do the 2 pull requests.
