Re: dm: enable opt-out of device-mapper dax support

[Dan Williams]

On Tue, Aug 1, 2017 at 12:45 PM, Dan Williams <dan.j.williams@xxxxxxxxx> wrote:
> On Tue, Aug 1, 2017 at 12:02 PM, Mike Snitzer <snitzer@xxxxxxxxxx> wrote:
>> On Tue, Aug 01 2017 at  2:12pm -0400,
>> Dan Williams <dan.j.williams@xxxxxxxxx> wrote:
>>
>>> Now that dax is no longer a default property of a block-device, i.e.
>>> ->direct_access() is not a block-device operation, we optionally enable
>>> device-mapper dax support with a new CONFIG_DM_DAX option.
>>>
>>> All the dax operations helpers are moved to a new file,
>>> drivers/md/dm-dax.c, that is optionally compiled when CONFIG_DM_DAX=y.
>>> Otherwise, we stub out all the operations with NULL function pointers
>>> and nop wrappers for the core dax routines.
>>>
>>> Cc: Alasdair Kergon <agk@xxxxxxxxxx>
>>> Cc: Mike Snitzer <snitzer@xxxxxxxxxx>
>>> Reported-by: Bart Van Assche <Bart.VanAssche@xxxxxxx>
>>> Signed-off-by: Dan Williams <dan.j.williams@xxxxxxxxx>
>>> ---
>>>  drivers/md/Kconfig     |   14 +++
>>>  drivers/md/Makefile    |    1
>>>  drivers/md/dm-dax.c    |  227 ++++++++++++++++++++++++++++++++++++++++++++++++
>>>  drivers/md/dm-dax.h    |   73 +++++++++++++++
>>>  drivers/md/dm-linear.c |   56 ------------
>>>  drivers/md/dm-snap.c   |    9 --
>>>  drivers/md/dm-stripe.c |   89 -------------------
>>>  drivers/md/dm-target.c |    7 -
>>>  drivers/md/dm.c        |  105 ++--------------------
>>>  drivers/md/dm.h        |   34 +++++++
>>>  10 files changed, 363 insertions(+), 252 deletions(-)
>>>  create mode 100644 drivers/md/dm-dax.c
>>>  create mode 100644 drivers/md/dm-dax.h
>>>
>>> diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig
>>> index 4a249ee86364..bf27b435f7cd 100644
>>> --- a/drivers/md/Kconfig
>>> +++ b/drivers/md/Kconfig
>>> @@ -200,7 +200,6 @@ config BLK_DEV_DM_BUILTIN
>>>  config BLK_DEV_DM
>>>       tristate "Device mapper support"
>>>       select BLK_DEV_DM_BUILTIN
>>> -     select DAX
>>>       ---help---
>>>         Device-mapper is a low level volume manager.  It works by allowing
>>>         people to specify mappings for ranges of logical sectors.  Various
>>> @@ -214,6 +213,19 @@ config BLK_DEV_DM
>>>
>>>         If unsure, say N.
>>>
>>> +config DM_DAX
>>> +     bool "Direct access (DAX) support"
>>> +     depends on BLK_DEV_DM
>>> +     default BLK_DEV_PMEM
>>> +     select DAX
>>> +     ---help---
>>> +       Enable DAX support for the device-mapper linear and stripe
>>> +       targets for use with DAX capable block devices like /dev/pmemN.
>>> +       If you have a DAX capable block device and have enabled
>>> +       filesystem DAX support (CONFIG_FS_DAX), then say Y.
>>> +
>>> +       If unsure, say N.
>>> +
>>
>> I'm questioning the need to have yet another Kbuild CONFIG option.  If
>> the user has enabled CONFIG_BLK_DEV_PMEM and CONFIG_FS_DAX (DAX already
>> gets selected by CONFIG_FS_DAX) then shouldn't the DM capabilities just
>> be enabled?
>>
>> Guess I'm just skeptical of: why do we want to move to a model where
>> users need to opt-in to DM support for DAX?
>>
>> I also _really_ don't like each target's DAX support being colocated in
>> drivers/md/dm-dax.c
>>
>> This all looks and feels like a serious step backwards.
>
> Ok, you want ifdef'd sections of DAX code in each target and make
> DM_DAX a silent option that gets enabled with BLK_DEV_PMEM, anything
> else?

Actually, no, I was thrown off by Bart's suggestion to move code
around. I can handle this all by deleting "select DAX" and adding more
stubbed out dax helpers.