Re: [Qemu-devel] [PATCH V6 0/6] nvdimm: support MAP_SYNC for memory-backend-file

[Yi Zhang <yi.z.zhang@linux.intel.com>]

On 2018-12-17 at 10:27:50 -0500, Michael S. Tsirkin wrote:
> On Mon, Dec 17, 2018 at 01:53:54PM +0800, Yi Zhang wrote:
> > On 2018-12-12 at 10:06:13 -0500, Michael S. Tsirkin wrote:
> > > On Wed, Dec 12, 2018 at 04:11:44PM +0800, Zhang Yi wrote:
> > > > Linux 4.15 introduces a new mmap flag MAP_SYNC, which can be used to
> > > > guarantee the write persistence to mmap'ed files supporting DAX (e.g.,
> > > > files on ext4/xfs file system mounted with '-o dax').
> > > > 
> > > > A description of MAP_SYNC and MAP_SHARED_VALIDATE can be found at
> > > >     https://patchwork.kernel.org/patch/10028151/
> > > > 
> > > > In order to make sure that the file metadata is in sync after a fault 
> > > > while we are writing a shared DAX supporting backend files, this
> > > > patch-set enables QEMU to use MAP_SYNC flag for memory-backend-dax-file.
> > > > 
> > > > As the DAX vs DMA truncated issue was solved, we refined the code and
> > > > send out this feature for the v5 version.
> > > > 
> > > > A new auto on/off option 'sync' is added to memory-backend-file:
> > > >  - on:  try to pass MAP_SYNC to mmap(2); if MAP_SYNC is not supported or
> > > >         'share=off', QEMU will abort
> > > >  - off: never pass MAP_SYNC to mmap(2)
> > > >  - auto (default): if MAP_SYNC is supported and 'share=on', work as if
> > > >         'sync=on'; otherwise, work as if 'sync=off'
> > > 
> > > Can this change be limited to a real DAX device?
> > > Then you won't need to bother with flags at all.
> > Thanks Micheal's review,
> > 
> > For real dax device, we still need to have a option to turn sync off.
> 
> Not sure why really. Why use an nvdimm if you don't actually
> want it to be persistent?
> 
> 
> > > 
> > > And am I mistaken in thinking that this will affect all
> > > guest memory with share=on? Or do I misunderstand?
> > 
> > Only file-backend memory with share=on.
> 
> Yes but that's everyone if e.g. vhost-user needs to be active.
> 
> So frankly I don't think it's a good idea to change the default like this
> unless it's limited to nvdimm in some way.

Yes, fine, I will limit this flag only on an nvdimm backend. Thanks,
Micheal.

> 
> > > 
> > > 
> > > 
> > > > Changes in v6:
> > > >  * Pankaj: 3/7 are squashed with 2/7
> > > >  * Pankaj: 7/7 update comments to "consistent filesystem metadata".
> > > >  * Pankaj, Igor: 1/7 Added Reviewed-by in patch-1/7
> > > >  * Stefan, 4/7 move the include header from "/linux/mman.h" to "osdep.h"
> > > >  * Stefan, 5/7 Add missing "munmap"
> > > >  * Stefan, 2/7 refine the shared/flag.
> > > > 
> > > > Changes in v5:
> > > >  * Add patch 1 to fix a memory leak issue.
> > > >  * Refine the patch 4-6
> > > >  * Remove the patch 3 as we already change the parameter from "shared" to
> > > >    "flags"
> > > > 
> > > > Changes in v4:
> > > >  * Add patch 1-3 to switch some functions to a single 'flags'
> > > >    parameters. (Michael S. Tsirkin)
> > > >  * v3 patch 1-3 become v4 patch 4-6.
> > > >  * Patch 4: move definitions of MAP_SYNC and MAP_SHARED_VALIDATE to a
> > > >    new header file under include/standard-headers/linux/. (Michael S. Tsirkin)
> > > >  * Patch 6: refine the description of the 'sync' option. (Michael S. Tsirkin)
> > > > 
> > > > Changes in v3:
> > > >  * Patch 1: add MAP_SHARED_VALIDATE in both sync=on and sync=auto
> > > >    cases, and add back the retry mechanism. MAP_SYNC will be ignored
> > > >    by Linux kernel 4.15 if MAP_SHARED_VALIDATE is missed.
> > > >  * Patch 1: define MAP_SYNC and MAP_SHARED_VALIDATE as 0 on non-Linux
> > > >    platforms in order to make qemu_ram_mmap() compile on those platforms.
> > > >  * Patch 2&3: include more information in error messages of
> > > >    memory-backend in hope to help user to identify the error.
> > > >    (Dr. David Alan Gilbert)
> > > >  * Patch 3: fix typo in the commit message. (Dr. David Alan Gilbert)
> > > > 
> > > > Changes in v2:
> > > >  * Add 'sync' option to control the use of MAP_SYNC. (Eduardo Habkost)
> > > >  * Remove the unnecessary set of MAP_SHARED_VALIDATE in some cases and
> > > >    the retry mechanism in qemu_ram_mmap(). (Michael S. Tsirkin)
> > > >  * Move OS dependent definitions of MAP_SYNC and MAP_SHARED_VALIDATE
> > > >    to osdep.h. (Michael S. Tsirkin)
> > > > 
> > > > Zhang Yi (6):
> > > >   numa: Fixed the memory leak of numa error message
> > > >   util/mmap-alloc: switch qemu_ram_mmap() to 'flags' parameter
> > > >   util/mmap-alloc: support MAP_SYNC in qemu_ram_mmap()
> > > >   util/mmap-alloc: Switch the RAM_SYNC flags to OnOffAuto
> > > >   hostmem: add more information in error messages
> > > >   hostmem-file: add 'sync' option
> > > > 
> > > >  backends/hostmem-file.c   | 45 +++++++++++++++++++++++++++++++++++++++++++--
> > > >  backends/hostmem.c        |  8 +++++---
> > > >  docs/nvdimm.txt           | 20 +++++++++++++++++++-
> > > >  exec.c                    |  9 +++++----
> > > >  include/exec/memory.h     | 18 ++++++++++++++++++
> > > >  include/exec/ram_addr.h   |  1 +
> > > >  include/qemu/mmap-alloc.h | 20 +++++++++++++++++++-
> > > >  include/qemu/osdep.h      | 29 +++++++++++++++++++++++++++++
> > > >  numa.c                    |  1 +
> > > >  qemu-options.hx           | 22 +++++++++++++++++++++-
> > > >  util/mmap-alloc.c         | 26 +++++++++++++++++++++-----
> > > >  util/oslib-posix.c        |  8 +++++++-
> > > >  12 files changed, 189 insertions(+), 18 deletions(-)
> > > > 
> > > > -- 
> > > > 2.7.4
>