| Message ID | 151996282448.28483.10415125852182473579.stgit@dwillia2-desk3.amr.corp.intel.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Headers | show |
On Thu, Mar 01, 2018 at 07:53:44PM -0800, Dan Williams wrote: > The current IS_DAX() helper that checks if a file is in DAX mode serves > two purposes. It is a control flow branch condition for DAX vs > non-DAX paths and it is a mechanism to perform dead code elimination. The > dead code elimination is required in the CONFIG_FS_DAX=n case since > there are symbols in fs/dax.c that will be elided. While the > dead code elimination can be addressed with nop stubs for the fs/dax.c > symbols that does not address the need for a DAX control flow helper > where fs/dax.c symbols are not involved. > > Moreover, the control flow changes, in some cases, need to be cognizant > of whether the DAX file is a typical file or a Device-DAX special file. > Introduce IS_DEVDAX() and IS_FSDAX() to simultaneously address the > file-type control flow and dead-code elimination use cases. IS_DAX() > will be deleted after all sites are converted to use the file-type > specific helper. > > Note, this change is also a pre-requisite for fixing the definition of > the S_DAX inode flag in the CONFIG_FS_DAX=n + CONFIG_DEV_DAX=y case. > The flag needs to be defined, non-zero, if either DAX facility is > enabled. > > Cc: "Theodore Ts'o" <tytso@mit.edu> > Cc: Andreas Dilger <adilger.kernel@dilger.ca> > Cc: Alexander Viro <viro@zeniv.linux.org.uk> > Cc: "Darrick J. Wong" <darrick.wong@oracle.com> > Cc: linux-xfs@vger.kernel.org > Cc: Matthew Wilcox <mawilcox@microsoft.com> > Cc: Ross Zwisler <ross.zwisler@linux.intel.com> > Cc: <stable@vger.kernel.org> > Fixes: dee410792419 ("/dev/dax, core: file operations and dax-mmap") > Reported-by: Jan Kara <jack@suse.cz> > Reviewed-by: Jan Kara <jack@suse.cz> > Signed-off-by: Dan Williams <dan.j.williams@intel.com> > --- > include/linux/fs.h | 22 ++++++++++++++++++++++ > 1 file changed, 22 insertions(+) > > diff --git a/include/linux/fs.h b/include/linux/fs.h > index 79c413985305..bd0c46880572 100644 > --- a/include/linux/fs.h > +++ b/include/linux/fs.h > @@ -1909,6 +1909,28 @@ static inline bool sb_rdonly(const struct super_block *sb) { return sb->s_flags > #define IS_WHITEOUT(inode) (S_ISCHR(inode->i_mode) && \ > (inode)->i_rdev == WHITEOUT_DEV) > > +static inline bool IS_DEVDAX(struct inode *inode) > +{ > + if (!IS_ENABLED(CONFIG_DEV_DAX)) > + return false; > + if ((inode->i_flags & S_DAX) == 0) > + return false; > + if (!S_ISCHR(inode->i_mode)) > + return false; > + return true; > +} > + > +static inline bool IS_FSDAX(struct inode *inode) > +{ > + if (!IS_ENABLED(CONFIG_FS_DAX)) > + return false; I echo Jan's complaint from the last round that the dead code elimination here is subtle, as compared to: #if IS_ENABLED(CONFIG_FS_DAX) static inline bool IS_FSDAX(struct inode *inode) { ... } #else # define IS_FSDAX(inode) (false) #endif But I guess even with that we're relying on dead code elimination higher up in the call stack... > + if ((inode->i_flags & S_DAX) == 0) > + return false; > + if (S_ISCHR(inode->i_mode)) > + return false; I'm curious, do we have character devices with S_DAX set? I /think/ we're expecting that only block/char devices and files will ever have S_DAX set, so IS_FSDAX is only true for block devices and files. Right? (A comment here about why S_ISCHR->false here would be helpful.) --D > + return true; > +} > + > static inline bool HAS_UNMAPPED_ID(struct inode *inode) > { > return !uid_valid(inode->i_uid) || !gid_valid(inode->i_gid); > > -- > To unsubscribe from this list: send the line "unsubscribe linux-xfs" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html
On Fri, Mar 2, 2018 at 9:45 AM, Darrick J. Wong <darrick.wong@oracle.com> wrote: > On Thu, Mar 01, 2018 at 07:53:44PM -0800, Dan Williams wrote: >> The current IS_DAX() helper that checks if a file is in DAX mode serves >> two purposes. It is a control flow branch condition for DAX vs >> non-DAX paths and it is a mechanism to perform dead code elimination. The >> dead code elimination is required in the CONFIG_FS_DAX=n case since >> there are symbols in fs/dax.c that will be elided. While the >> dead code elimination can be addressed with nop stubs for the fs/dax.c >> symbols that does not address the need for a DAX control flow helper >> where fs/dax.c symbols are not involved. >> >> Moreover, the control flow changes, in some cases, need to be cognizant >> of whether the DAX file is a typical file or a Device-DAX special file. >> Introduce IS_DEVDAX() and IS_FSDAX() to simultaneously address the >> file-type control flow and dead-code elimination use cases. IS_DAX() >> will be deleted after all sites are converted to use the file-type >> specific helper. >> >> Note, this change is also a pre-requisite for fixing the definition of >> the S_DAX inode flag in the CONFIG_FS_DAX=n + CONFIG_DEV_DAX=y case. >> The flag needs to be defined, non-zero, if either DAX facility is >> enabled. >> >> Cc: "Theodore Ts'o" <tytso@mit.edu> >> Cc: Andreas Dilger <adilger.kernel@dilger.ca> >> Cc: Alexander Viro <viro@zeniv.linux.org.uk> >> Cc: "Darrick J. Wong" <darrick.wong@oracle.com> >> Cc: linux-xfs@vger.kernel.org >> Cc: Matthew Wilcox <mawilcox@microsoft.com> >> Cc: Ross Zwisler <ross.zwisler@linux.intel.com> >> Cc: <stable@vger.kernel.org> >> Fixes: dee410792419 ("/dev/dax, core: file operations and dax-mmap") >> Reported-by: Jan Kara <jack@suse.cz> >> Reviewed-by: Jan Kara <jack@suse.cz> >> Signed-off-by: Dan Williams <dan.j.williams@intel.com> >> --- >> include/linux/fs.h | 22 ++++++++++++++++++++++ >> 1 file changed, 22 insertions(+) >> >> diff --git a/include/linux/fs.h b/include/linux/fs.h >> index 79c413985305..bd0c46880572 100644 >> --- a/include/linux/fs.h >> +++ b/include/linux/fs.h >> @@ -1909,6 +1909,28 @@ static inline bool sb_rdonly(const struct super_block *sb) { return sb->s_flags >> #define IS_WHITEOUT(inode) (S_ISCHR(inode->i_mode) && \ >> (inode)->i_rdev == WHITEOUT_DEV) >> >> +static inline bool IS_DEVDAX(struct inode *inode) >> +{ >> + if (!IS_ENABLED(CONFIG_DEV_DAX)) >> + return false; >> + if ((inode->i_flags & S_DAX) == 0) >> + return false; >> + if (!S_ISCHR(inode->i_mode)) >> + return false; >> + return true; >> +} >> + >> +static inline bool IS_FSDAX(struct inode *inode) >> +{ >> + if (!IS_ENABLED(CONFIG_FS_DAX)) >> + return false; > > I echo Jan's complaint from the last round that the dead code > elimination here is subtle, as compared to: > > #if IS_ENABLED(CONFIG_FS_DAX) > static inline bool IS_FSDAX(struct inode *inode) { ... } > #else > # define IS_FSDAX(inode) (false) > #endif > > But I guess even with that we're relying on dead code elimination higher > up in the call stack... If IS_FSDAX() was only a dead-code elimination mechanism rather than a runtime branch condition then I agree. Otherwise I think IS_ENABLED() is suitable and not subtle, especially when used in a header file. >> + if ((inode->i_flags & S_DAX) == 0) >> + return false; >> + if (S_ISCHR(inode->i_mode)) >> + return false; > > I'm curious, do we have character devices with S_DAX set? Yes, Device-DAX, see: ab68f2622136 /dev/dax, pmem: direct access to persistent memory > I /think/ we're expecting that only block/char devices and files will > ever have S_DAX set, so IS_FSDAX is only true for block devices and > files. Right? We had S_DAX on block-devices for a short while, but deleted it and went with the Device-DAX interface instead. So it's only regular files and /dev/daxX.Y nodes these days. > (A comment here about why S_ISCHR->false here would be helpful.) Ok.
> +static inline bool IS_DEVDAX(struct inode *inode) > +{ > + if (!IS_ENABLED(CONFIG_DEV_DAX)) > + return false; > + if ((inode->i_flags & S_DAX) == 0) > + return false; > + if (!S_ISCHR(inode->i_mode)) > + return false; > + return true; > +} > + > +static inline bool IS_FSDAX(struct inode *inode) > +{ > + if (!IS_ENABLED(CONFIG_FS_DAX)) > + return false; > + if ((inode->i_flags & S_DAX) == 0) > + return false; > + if (S_ISCHR(inode->i_mode)) > + return false; > + return true; Encoding the is char device or not thing here is just nasty. I think this is going entirely in the wrong direction.
diff --git a/include/linux/fs.h b/include/linux/fs.h index 79c413985305..bd0c46880572 100644 --- a/include/linux/fs.h +++ b/include/linux/fs.h @@ -1909,6 +1909,28 @@ static inline bool sb_rdonly(const struct super_block *sb) { return sb->s_flags #define IS_WHITEOUT(inode) (S_ISCHR(inode->i_mode) && \ (inode)->i_rdev == WHITEOUT_DEV) +static inline bool IS_DEVDAX(struct inode *inode) +{ + if (!IS_ENABLED(CONFIG_DEV_DAX)) + return false; + if ((inode->i_flags & S_DAX) == 0) + return false; + if (!S_ISCHR(inode->i_mode)) + return false; + return true; +} + +static inline bool IS_FSDAX(struct inode *inode) +{ + if (!IS_ENABLED(CONFIG_FS_DAX)) + return false; + if ((inode->i_flags & S_DAX) == 0) + return false; + if (S_ISCHR(inode->i_mode)) + return false; + return true; +} + static inline bool HAS_UNMAPPED_ID(struct inode *inode) { return !uid_valid(inode->i_uid) || !gid_valid(inode->i_gid);
The current IS_DAX() helper that checks if a file is in DAX mode serves two purposes. It is a control flow branch condition for DAX vs non-DAX paths and it is a mechanism to perform dead code elimination. The dead code elimination is required in the CONFIG_FS_DAX=n case since there are symbols in fs/dax.c that will be elided. While the dead code elimination can be addressed with nop stubs for the fs/dax.c symbols that does not address the need for a DAX control flow helper where fs/dax.c symbols are not involved. Moreover, the control flow changes, in some cases, need to be cognizant of whether the DAX file is a typical file or a Device-DAX special file. Introduce IS_DEVDAX() and IS_FSDAX() to simultaneously address the file-type control flow and dead-code elimination use cases. IS_DAX() will be deleted after all sites are converted to use the file-type specific helper. Note, this change is also a pre-requisite for fixing the definition of the S_DAX inode flag in the CONFIG_FS_DAX=n + CONFIG_DEV_DAX=y case. The flag needs to be defined, non-zero, if either DAX facility is enabled. Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Andreas Dilger <adilger.kernel@dilger.ca> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: "Darrick J. Wong" <darrick.wong@oracle.com> Cc: linux-xfs@vger.kernel.org Cc: Matthew Wilcox <mawilcox@microsoft.com> Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Cc: <stable@vger.kernel.org> Fixes: dee410792419 ("/dev/dax, core: file operations and dax-mmap") Reported-by: Jan Kara <jack@suse.cz> Reviewed-by: Jan Kara <jack@suse.cz> Signed-off-by: Dan Williams <dan.j.williams@intel.com> --- include/linux/fs.h | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+)