On 07/22/2014 10:47 PM, Matthew Wilcox wrote:
> One of the primary uses for NV-DIMMs is to expose them as a block device
> and use a filesystem to store files on the NV-DIMM. While that works,
> it currently wastes memory and CPU time buffering the files in the page
> cache. We have support in ext2 for bypassing the page cache, but it
> has some races which are unfixable in the current design. This series
> of patches rewrite the underlying support, and add support for direct
> access to ext4.
>
> I would particularly welcome feedback from mm people on patch 5 ("Add
> vm_replace_mixed()") and from fs people on patch 7 ("Add copy_to_iter(),
> copy_from_iter() and iov_iter_zero()").
>
> This iteration of the patchset rebases to 3.16-rc6 and makes substantial
> changes based on Jan Kara's feedback:
>
Hi Matthew
Thank you for all your hard work. I will review and test this series, and
report.
Have you please pushed this tree to git hub. It used to be on the prd
tree, if you could just add another branch there, it would be cool.
(https://github.com/01org/prd)
I've been testing performance with the 3.14 based tree against a post-pmfs
fs that uses prd, vs ext4-dax vs shmem: (Same memory workset sizes)
ext4-dax is over prd same as pmfs, 32G + 32G system memory
shmem is using the same 32G work-sets but on a 64G system memory (no prd)
With reads they are all the same with pmfs-prd a bit on top.
But writes, shmem is by far far better. I'll Investigate this farther, it
does not make sense looks like a serialization of writes somewhere.
(All numbers rounded)
fio 4K random reads 24 threads (different files):
pmfs-prd: 7,300,000
ext4-dax: 6,200,000
shmem: 6,150,000
fio 4k random writes 24 threads (different files):
pmfs-prd: 900,000
ext4-dax: 800,000
shmem: 3,500,000
BTW:
With the new pmfs-prd which I have also ported to the DAX tree - the old
pmfs was copy/pasting xip-write and xip-mmap from generic code with some
private hacks, but was still using xip-read stuff. I have just moved the
old xip-read stuff inside (like write path did before) and so ported to the
dax tree which removed these functions. I have spent a couple of hours on
looking on the DAX read/write/mmap path vs the current pmfs read/write/mmap
paths. Though they are very similar and clearly originated from the same code
The new DAX code is even more alien to pmfs, specially with the current use
of buffer-heads. (It could be made to work but with lots of extra shimming)
So the current DAX form is more "block" based and not very suited to a pure
memory based FS. If any of those come up they will need their own interfaces
more similar to shmem's private implementation.
Above you said regarding the old xip:
".. but it has some races which are unfixable in the current design"
I have tested pmfs-prd under xfstest and I'm passing lots of them. I cannot
easily spot the race you are talking about. (More like dir seeks and missing
stuff) Could you point me to a test I should run, to try and find it?
Thanks have a good day
Boaz
> - Handles errors in copy_user_bh()
> - Changes calling convention for dax_get_addr() / dax_get_pfn() to take a
> blkbits argument instead of an inode argument
> - Cache inode->i_blkbits in a local variable
> - Rename file offset to 'pos' to fit the rest of the VFS
> - Added a FIXME to fall back to buffered I/O if the filesystem doesn't
> support filling a hole from within the direct I/O path. Mysterious
> and complex are the ways of get_block implementations.
> - Moved the call to inode_dio_done() to after end_io() to fix a race
> - Added a comment about why we have to recheck i_size under the page lock
> - Use vm_insert_page() in the COW path instead of returning VM_FAULT_COWED
> - Handle errors coming back from dax_get_pfn() correctly in do_dax_fault()
> - Removes zero pages from the process' address space before trying to
> replace them with the PFN of the newly allocated block
> - Factor out bdev_direct_access() to support partitioning properly
> - Rework the i_mmap_mutex locking to remove an inversion vs the page lock
> - Make the ext2 rename of -o xip to -o dax more graceful
> - Only update file mtime/ctime on a write fault, not a read fault
>
>
> Matthew Wilcox (21):
> Fix XIP fault vs truncate race
> Allow page fault handlers to perform the COW
> axonram: Fix bug in direct_access
> Change direct_access calling convention
> Add vm_replace_mixed()
> Introduce IS_DAX(inode)
> Add copy_to_iter(), copy_from_iter() and iov_iter_zero()
> Replace XIP read and write with DAX I/O
> Replace ext2_clear_xip_target with dax_clear_blocks
> Replace the XIP page fault handler with the DAX page fault handler
> Replace xip_truncate_page with dax_truncate_page
> Replace XIP documentation with DAX documentation
> Remove get_xip_mem
> ext2: Remove ext2_xip_verify_sb()
> ext2: Remove ext2_use_xip
> ext2: Remove xip.c and xip.h
> Remove CONFIG_EXT2_FS_XIP and rename CONFIG_FS_XIP to CONFIG_FS_DAX
> ext2: Remove ext2_aops_xip
> Get rid of most mentions of XIP in ext2
> xip: Add xip_zero_page_range
> brd: Rename XIP to DAX
>
> Ross Zwisler (1):
> ext4: Add DAX functionality
>
> Documentation/filesystems/Locking | 3 -
> Documentation/filesystems/dax.txt | 91 +++++++
> Documentation/filesystems/ext4.txt | 2 +
> Documentation/filesystems/xip.txt | 68 -----
> arch/powerpc/sysdev/axonram.c | 14 +-
> drivers/block/Kconfig | 13 +-
> drivers/block/brd.c | 22 +-
> drivers/s390/block/dcssblk.c | 19 +-
> fs/Kconfig | 21 +-
> fs/Makefile | 1 +
> fs/block_dev.c | 28 +++
> fs/dax.c | 503 +++++++++++++++++++++++++++++++++++++
> fs/exofs/inode.c | 1 -
> fs/ext2/Kconfig | 11 -
> fs/ext2/Makefile | 1 -
> fs/ext2/ext2.h | 10 +-
> fs/ext2/file.c | 45 +++-
> fs/ext2/inode.c | 38 +--
> fs/ext2/namei.c | 13 +-
> fs/ext2/super.c | 53 ++--
> fs/ext2/xip.c | 91 -------
> fs/ext2/xip.h | 26 --
> fs/ext4/ext4.h | 6 +
> fs/ext4/file.c | 53 +++-
> fs/ext4/indirect.c | 18 +-
> fs/ext4/inode.c | 51 ++--
> fs/ext4/namei.c | 10 +-
> fs/ext4/super.c | 39 ++-
> fs/open.c | 5 +-
> include/linux/blkdev.h | 6 +-
> include/linux/fs.h | 49 +++-
> include/linux/mm.h | 9 +-
> include/linux/uio.h | 3 +
> mm/Makefile | 1 -
> mm/fadvise.c | 6 +-
> mm/filemap.c | 6 +-
> mm/filemap_xip.c | 483 -----------------------------------
> mm/iov_iter.c | 237 +++++++++++++++--
> mm/madvise.c | 2 +-
> mm/memory.c | 45 ++--
> 40 files changed, 1228 insertions(+), 875 deletions(-)
> create mode 100644 Documentation/filesystems/dax.txt
> delete mode 100644 Documentation/filesystems/xip.txt
> create mode 100644 fs/dax.c
> delete mode 100644 fs/ext2/xip.c
> delete mode 100644 fs/ext2/xip.h
> delete mode 100644 mm/filemap_xip.c
>
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
