[lkml]   [2006]   [May]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
    SubjectRe: [PATCH 1/3] Vectorize aio_read/aio_write methods
    On Tue, May 09, 2006 at 12:13:05PM -0700, Andrew Morton wrote:
    > > there's another patch ontop which I didn't bother to redo until this is
    > > accepted which kills a lot more code. After that filesystems only have
    > > to implement one method each for all kinds of read/write calls. Which
    > > allows to both make the mm/filemap.c far less complex and actually
    > > understandable aswell as for any filesystem that uses more complex
    > > read/write variants than direct filemap.c calls. In addition to these
    > > simplification we also get a feature (async vectored I/O) for free.
    > Fair enough, thanks. Simplifying filemap.c would be a win.
    > I'll crunch on these three patches in the normal fashion. It'll be good if
    > we can get the followup patch done within the next week or two so we can
    > get it all tested at the same time. Although from your description it
    > doesn't sound like it'll be completely trivial...

    That patch is lots of tirival and boring work. If anyone wants to beat
    me to it:

    - in any filesystem that implements the generic_file_aio_{read,write}
    directly remove these apply this patch to the file_operations

    - .read = generic_file_read,
    - .write = generic_file_write,
    + .read = do_sync_read,
    + .write = do_sync_write,

    Note that this does _not_ cause additional indirection for normal
    sys_read/sys_write calls because they call .aio_read/.aio_write
    directly. It's only needed because we have various places in the
    tree that like to call .read/.write directly

    - in the filesystems that implement more or less trivial wrappers
    around generic_file_read/generic_file_write to the
    aio_read/aio_write prototypes so they can set .read/.write as above

    - after that generic_file_read/generic_file_write/generic_file_read/
    generic_file_write_nolock should have no callers left and the code
    for read/write in mm/filemap.c can be collapsed into very few functions.
    What's left should be something like:

    - generic_file_aio_read
    (__generic_file_aio_read and generic_file_aio_read merged into one)
    - __generic_file_aio_write
    (basically the current __generic_file_aio_write_nolock)
    - generic_file_aio_write_nolock
    - generic_file_aio_write
    (small wrappers around __generic_file_aio_write)
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to
    More majordomo info at
    Please read the FAQ at

     \ /
      Last update: 2006-05-09 21:23    [W:0.021 / U:2.312 seconds]
    ©2003-2017 Jasper Spaans. hosted at Digital OceanAdvertise on this site