New topic branch for block + gup work?

New topic branch for block + gup work?

[John Hubbard]

Hi Jens,

After you suggested a topic branch [1] as a way to address the recent
bio_map_user_iov() conflict in linux-next, I've reviewed a few more
patchsets in mm, and am now starting to suspect that a topic branch
would be ideal here.

Logan's "Userspace P2PDMA with O_DIRECT NVMe devices" series [2], my
"convert most filesystems to pin_user_pages_fast()" series [3], and the
block layer change from [1], all conflict in iov_iter*, and in
bio_map_user_iov().

Less of an issue but still worth considering, Dan's "Fix the DAX-gup
mistake" series [4] conflicts in gup.c, too.

Maybe:

    gup_bio

, or something like that, as a topic branch?

Everyone: thoughts, preferences here?


[1] https://lore.kernel.org/r/20220901161722.739d2013@canb.auug.org.au

[2] https://lore.kernel.org/r/20220825152425.6296-1-logang@deltatee.com

[3] https://lore.kernel.org/r/20220831041843.973026-1-jhubbard@nvidia.com

[4] https://lore.kernel.org/r/166225775968.2351842.11156458342486082012.stgit@dwillia2-xfh.jf.intel.com


thanks,

-- 
John Hubbard
NVIDIA

Re: New topic branch for block + gup work?

[Jens Axboe]

On 9/5/22 5:16 PM, John Hubbard wrote:
> Hi Jens,
> 
> After you suggested a topic branch [1] as a way to address the recent
> bio_map_user_iov() conflict in linux-next, I've reviewed a few more
> patchsets in mm, and am now starting to suspect that a topic branch
> would be ideal here.
> 
> Logan's "Userspace P2PDMA with O_DIRECT NVMe devices" series [2], my
> "convert most filesystems to pin_user_pages_fast()" series [3], and the
> block layer change from [1], all conflict in iov_iter*, and in
> bio_map_user_iov().
> 
> Less of an issue but still worth considering, Dan's "Fix the DAX-gup
> mistake" series [4] conflicts in gup.c, too.
> 
> Maybe:
> 
>     gup_bio
> 
> , or something like that, as a topic branch?
> 
> Everyone: thoughts, preferences here?

My suggestion would be to branch from for-6.1/block, then we can
apply the gup patches on top of that. I'd probably just call it
for-6.1/block-gup.

-- 
Jens Axboe