mbox series

[PATCHSET,v2,0/10] Move away from remap_pfn_range()

Message ID 20240327191933.607220-1-axboe@kernel.dk (mailing list archive)
Headers show
Series Move away from remap_pfn_range() | expand

Message

Jens Axboe March 27, 2024, 7:13 p.m. UTC 
Hi,

This series switches both the ring, sqes, and kbuf side away from
using remap_pfn_range(). It's a v2 of the one posted here:

https://lore.kernel.org/io-uring/20240321144831.58602-1-axboe@kernel.dk/

and includes the ring/sqes side as well as the first few patches, with
the idea being that we could move those to stable as well to solve the
problem of fragmented memory in production systems not being able to
initialize bigger rings.

This series has been co-developed with Pavel Begunkov.

Patch 1 is just a prep patch, and patches 2-3 add ring support, and
patch 4 just unifies some identical code.

Patches 5-7 cleanup some kbuf side code, and patch 8 prepares buffer
lists to be reference counted, and then patch 9 can finally switch
kbuf to also use the nicer vm_insert_pages().

With this, no more remap_pfn_range(), and no more manual cleanup of
having used it.

 include/linux/io_uring_types.h |   4 -
 io_uring/io_uring.c            | 261 +++++++++++++++++++----------
 io_uring/io_uring.h            |   8 +-
 io_uring/kbuf.c                | 290 ++++++++++-----------------------
 io_uring/kbuf.h                |  11 +-
 io_uring/rsrc.c                |  36 ----
 mm/nommu.c                     |   7 +
 7 files changed, 277 insertions(+), 340 deletions(-)