[RTEMS Project] #3894: Replace the device filesystem with a specialization of the IMFS

[RTEMS trac]

#3894: Replace the device filesystem with a specialization of the IMFS
-----------------------------+------------------------------
 Reporter:  Sebastian Huber  |       Owner:  Sebastian Huber
     Type:  enhancement      |      Status:  assigned
 Priority:  normal           |   Milestone:  5.1
Component:  fs               |     Version:  5
 Severity:  normal           |  Resolution:
 Keywords:                   |  Blocked By:
 Blocking:  3898             |
-----------------------------+------------------------------

Comment (by Sebastian Huber):

 Replying to [comment:2 Sebastian Huber]:
 > Replying to [comment:1 Chris Johns]:
 > > How doe the size of this approach compare to the `devfs` approach?
 >
 > I don't have numbers yet. I guess it will slightly increase.

 I checked the code size change of the devfs04 test before and after the
 change to the IMFS specialization:

 size pre/devfs04.exe
    text    data     bss     dec     hex filename
   57328    1152   17312   75792   12810 pre/devfs04.exe

 size post/devfs04.exe
    text    data     bss     dec     hex filename
   58384    1152   16096   75632   12770 post/devfs04.exe

 The code size increases by 1056 bytes on SPARCV8.

 In case the support for legacy IO drivers is disabled via

 {{{
 #define CONFIGURE_IMFS_DISABLE_MKNOD_DEVICE
 #define CONFIGURE_IMFS_DISABLE_MKNOD
 }}}

 the code size drops to:
 size devfs04.exe
    text    data     bss     dec     hex filename
   56496    1152   16096   73744   12010 devfs04.exe

 Using IMFS_add_node() gives you access to all file handlers and removes
 four layers of indirection, e.g.

 read() -> device_read() -> rtems_deviceio_read() -> rtems_io_read() ->
 _IO_Driver_address_table -> device handler

 compared to

 read() -> device handler.

--
Ticket URL: <http://devel.rtems.org/ticket/3894#comment:9>
RTEMS Project <http://www.rtems.org/>
RTEMS Project