Re: linux-next: build failure after merge of the device-mapper tree

[Jens Axboe]

On 5/31/18 8:22 AM, Jens Axboe wrote:
> On 5/31/18 8:20 AM, Jens Axboe wrote:
>> On 5/31/18 6:10 AM, Mike Snitzer wrote:
>>> On Thu, May 31 2018 at 12:35am -0400,
>>> Jens Axboe <axboe@xxxxxxxxx> wrote:
>>>
>>>> On May 30, 2018, at 10:23 PM, Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx> wrote:
>>>>>
>>>>> Hi all,
>>>>>
>>>>> After merging the device-mapper tree, today's linux-next build (x86_64
>>>>> allmodconfig) failed like this:
>>>>>
>>>>> drivers/md/dm-writecache.c: In function 'writecache_dtr':
>>>>> drivers/md/dm-writecache.c:1799:3: error: implicit declaration of function 'bioset_free'; did you mean 'bvec_free'? [-Werror=implicit-function-declaration]
>>>>>   bioset_free(wc->bio_set);
>>>>>   ^~~~~~~~~~~
>>>>>   bvec_free
>>>>> drivers/md/dm-writecache.c: In function 'writecache_ctr':
>>>>> drivers/md/dm-writecache.c:1929:17: error: implicit declaration of function 'bioset_create'; did you mean 'bioset_exit'? [-Werror=implicit-function-declaration]
>>>>>   wc->bio_set = bioset_create(BIO_POOL_SIZE,
>>>>>                 ^~~~~~~~~~~~~
>>>>>                 bioset_exit
>>>>> drivers/md/dm-writecache.c:1929:15: warning: assignment makes pointer from integer without a cast [-Wint-conversion]
>>>>>   wc->bio_set = bioset_create(BIO_POOL_SIZE,
>>>>>               ^
>>>>>
>>>>> Caused by commit
>>>>>
>>>>>  2105231db61b ("dm: add writecache target")
>>>>>
>>>>> interacting with commit
>>>>>
>>>>>  dad08527525f ("block: Drop bioset_create()")
>>>>>
>>>>> from the block tree.
>>>>>
>>>>> Can we please consider not immediately dropping APIs if at all
>>>>> possible. :-(
>>>>
>>>> Iâll drop the last patch, we can do that at the end of the merge window instead. 
>>>>
>>>
>>> FYI, I've since updated dm-writecache to use the new APIs
>>
>> Thanks Mike. I've rebased for-4.18/block to fold a bug fix for the
>> bounce code, and drop the patch that gets rid of the old API. When
>> we get to the end of the merge window, I'll push that last patch.
>> That's much more sensible, how it should have been done from the
>> get-go.
> 
> Actually, maybe I should roll them back, since you both fixed that
> up AND based on my tree - which is of course totally fine, but
> now I'm thinking the rebase will do more harm than good in this
> case.

Talked to Mike offline, but for Stephen's sake I'll reiterate here
too. Since I'm assuming that dm-writecache is the only problematic
user here, and since Mike based dm-4.18 on the block tree, I considered
it saner to leave things as-is. This means that the API removal is still
in the tree.

Stephen, if you see further conflicts due to that, do let me know and
I'll do a proper revert of the patch instead. Hopefully it'll be smooth
sailing from here on out.

-- 
Jens Axboe