Messages in this thread | | | Date | Sat, 24 May 2008 00:51:33 +0200 | From | Jan Kara <> | Subject | Re: [PATCH] VFS: Pagecache usage optimization on pagesize !=blocksize environment |
| |
> On Thu, 22 May 2008 16:31:15 +0900 Hisashi Hifumi <hifumi.hisashi@oss.ntt.co.jp> wrote: > > One other thing we should think about here is the `nobh' mode which the > extX filesystems support (although I have a feeling that Nick might > have broken this ;)) We also have data=ordered, data=writeback and > data=journal to think about. This optimisation might not be > appropriate at all to data=journal mode, but I haven't looked into > that. Why do you think so? We mess with buffer_dirty bits in data=journal mode but as far as I understand the patch, it only does not read the page if the part we need is marked as uptodate in buffers. And this should be safe. But I'm slightly confused that the patch helps because I've always thought that mpage_readpage() (which is what we end up calling from do_generic_mapping_read()) always reads the whole page. Thus either all buffers in the page or none of them are uptodate... So what do I miss here?
Honza -- Jan Kara <jack@suse.cz> SuSE CR Labs
| |