Re: [PATCH RFC v3 for-6.8/block 02/17] xen/blkback: use bdev api in xen_update_blkif_status()

[Jan Kara]

Hi Kuai!

On Thu 04-01-24 20:19:05, Yu Kuai wrote:
> 在 2024/01/04 19:06, Jan Kara 写道:
> > On Thu 21-12-23 16:56:57, Yu Kuai wrote:
> > > From: Yu Kuai <yukuai3@xxxxxxxxxx>
> > > 
> > > Avoid to access bd_inode directly, prepare to remove bd_inode from
> > > block_devcie.
> > > 
> > > Signed-off-by: Yu Kuai <yukuai3@xxxxxxxxxx>
> > > ---
> > >   drivers/block/xen-blkback/xenbus.c | 3 +--
> > >   1 file changed, 1 insertion(+), 2 deletions(-)
> > > 
> > > diff --git a/drivers/block/xen-blkback/xenbus.c b/drivers/block/xen-blkback/xenbus.c
> > > index e34219ea2b05..e645afa4af57 100644
> > > --- a/drivers/block/xen-blkback/xenbus.c
> > > +++ b/drivers/block/xen-blkback/xenbus.c
> > > @@ -104,8 +104,7 @@ static void xen_update_blkif_status(struct xen_blkif *blkif)
> > >   		xenbus_dev_error(blkif->be->dev, err, "block flush");
> > >   		return;
> > >   	}
> > > -	invalidate_inode_pages2(
> > > -			blkif->vbd.bdev_handle->bdev->bd_inode->i_mapping);
> > > +	invalidate_bdev(blkif->vbd.bdev_handle->bdev);
> > 
> > This function uses invalidate_inode_pages2() while invalidate_bdev() ends
> > up using mapping_try_invalidate() and there are subtle behavioral
> > differences between these two (for example invalidate_inode_pages2() tries
> > to clean dirty pages using the ->launder_folio method). So I think you'll
> > need helper like invalidate_bdev2() for this.
> 
> Thanks for reviewing this patch, I know the differenct between then,
> what I don't understand is that why using invalidate_inode_pages2()
> here.

Well, then the change in behavior should be at least noted in the
changelog.

> sync_blockdev() is just called and 0 is returned, I think in this
> case it's safe to call invalidate_bdev() directly, or am I missing
> other things?

I still think there's a difference. invalidate_inode_pages2() also unmaps
memory mappings which mapping_try_invalidate() does not do. That being said
in xen_update_blkif_status() we seem to be bringing up a virtual block
device so before this function is called, anybody would have hard time
using anything in it. But this definitely needs a confirmation from Xen
maintainers and a good documentation of the behavioral change in the
changelog.

								Honza
-- 
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR