On Thu, Sep 19, 2019 at 05:11:28AM +0800, Gao Xiang wrote: > Hi Christoph, > > On Wed, Sep 18, 2019 at 11:24:12AM -0700, Christoph Hellwig wrote: > > Just as a note of record: I don't think either file system move > > is a good idea. erofs still needs a lot of work, especially in > > interacting with the mm code like abusing page->mapping. > > I know what you mean, that is a known stuff in the TODO list, > Z_EROFS_MAPPING_STAGING page->mapping just be used as a temporary > page mark since page->private is already pointed to another > structure, It's now be marked as an non-movable and anon pseudo > mapping and most mm code just skip this case. Add a word, these pages are all non-LRU and short lifetime temporary pages (and need to differentiate with other NULL mapping pages [a lot of different type pages could have this intermediate state]). Alternate way is to use some page flag but that is what I really want to avoid this limited resource. For EROFS, it's widely deployment on each new HUAWEI mobile phones on the market this year and all old HUAWEI modile phones are still supported by upgrading to EROFS version and there are many pending new features for EROFS and a mature fixed-sized output compression subsystem in the future if more fs users have interested in that and I think it's good for whole linux ecosystem not just on a single filesystem upstreaming basis and we will continue working on this area. Thanks, Gao Xiang > > I think a better way is to use a real address_space structure for > page->mapping to point. It's easy to update but I need some time > to verify. If I am wrong, please point it out... > > Thanks, > Gao Xiang > _______________________________________________ devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxx http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
