On Mon, Nov 28, 2016 at 2:51 PM, Ross Zwisler
<ross.zwisler@xxxxxxxxxxxxxxx> wrote:
> On Mon, Nov 28, 2016 at 02:13:29PM -0800, Dan Williams wrote:
>> On Mon, Nov 28, 2016 at 1:24 PM, Ross Zwisler
>> <ross.zwisler@xxxxxxxxxxxxxxx> wrote:
>> > On Mon, Nov 28, 2016 at 10:12:17PM +0100, Arnd Bergmann wrote:
>> >> Without the get_block based I/O, we get warnings when CONFIG_FS_IOMAP
>> >> is disabled:
>> >>
>> >> fs/dax.c:736:12: error: ‘dax_insert_mapping’ defined but not used [-Werror=unused-function]
>> >> fs/dax.c:512:12: error: ‘copy_user_dax’ defined but not used [-Werror=unused-function]
>> >> fs/dax.c:490:12: error: ‘dax_load_hole’ defined but not used [-Werror=unused-function]
>> >> fs/dax.c:294:14: error: ‘grab_mapping_entry’ defined but not used [-Werror=unused-function]
>> >>
>> >> This patch blindly marks those as __maybe_unused, which avoids the warnings.
>> >> However, I suspect that there is actually more code in this file that should
>> >> not be provided without CONFIG_FS_IOMAP even though we don't get a warning
>> >> for it, and that we actually want a different rework, so please treat this
>> >> as a bug report. I have applied the patch locally in my randconfig build
>> >> setup to avoid seeing the warnings.
>> >>
>> >> Fixes: 5ac65736f740 ("dax: rip out get_block based IO support")
>> >> Signed-off-by: Arnd Bergmann <arnd@xxxxxxxx>
>> >
>> > Thanks for the report. I think the right way to deal with this is to just
>> > select FS_IOMAP when we pull in the DAX code. I sent out a patch last week
>> > that does this:
>> >
>> > https://lkml.org/lkml/2016/11/23/591
>>
>> It seems awkward for both filesystems and the FS_DAX core to be
>> selecting FS_IOMAP. In the end FS_DAX and FS_IOMAP are both libraries
>> of functionality that a filesystem can optionally use. I think the
>> longer term FS_DAX stops being an independent user visible setting and
>> is instead selected by filesystems that want DAX.
>
> This doesn't make sense to me. DAX is a user-selectable option that changes
> behavior (at the user's request), but FS_IOMAP is a library of functionality
> that is required for XFS and for DAX.
>
> The filesystems can all work fine without DAX (hence the user option), but DAX
> and XFS at least require FS_IOMAP to behave correctly.
>
> If you made DAX a FS selectable option instead of a user selectable one, when
> would a FS know it needs to include DAX support?
With a user-selectable DAX knob per-filesystem, XFS_DAX, EXT4_DAX, etc...
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
