mbox series

[v5,0/8] Introduce cpu_dcache_is_aliasing() to fix DAX regression

Message ID 20240212163101.19614-1-mathieu.desnoyers@efficios.com
Headers
Series Introduce cpu_dcache_is_aliasing() to fix DAX regression |

Message

Mathieu Desnoyers Feb. 12, 2024, 4:30 p.m. UTC 
  This commit introduced in v4.0 prevents building FS_DAX on 32-bit ARM,
even on ARMv7 which does not have virtually aliased data caches:

commit d92576f1167c ("dax: does not work correctly with virtual aliasing caches")

It used to work fine before: I have customers using DAX over pmem on
ARMv7, but this regression will likely prevent them from upgrading their
kernel.

The root of the issue here is the fact that DAX was never designed to
handle virtually aliasing data caches (VIVT and VIPT with aliasing data
cache). It touches the pages through their linear mapping, which is not
consistent with the userspace mappings with virtually aliasing data
caches.

This patch series introduces cpu_dcache_is_aliasing() with the new
Kconfig option ARCH_HAS_CPU_CACHE_ALIASING and implements it for all
architectures. The implementation of cpu_dcache_is_aliasing() is either
evaluated to a constant at compile-time or a runtime check, which is
what is needed on ARM.

With this we can basically narrow down the list of architectures which
are unsupported by DAX to those which are really affected.

Testing done so far:

- Compile allyesconfig on x86-64,
- Compile allyesconfig on x86-64, with FS_DAX=n.
- Compile allyesconfig on x86-64, with DAX=n.
- Boot test after modifying alloc_dax() to force returning -EOPNOTSUPP
  even on x86-64, thus simulating the behavior expected on an
  architecture with data cache aliasing.

There are many more axes to test however. I would welcome Tested-by for:

- affected architectures,
- affected drivers,
- affected filesytems.

[ Based on commit "nvdimm/pmem: Fix leak on dax_add_host() failure". ]

Thanks,

Mathieu

Changes since v4:
- Move the change which makes alloc_dax() return ERR_PTR(-EOPNOTSUPP)
  when CONFIG_DAX=n earlier in the series,
- Fold driver cleanup patches into their respective per-driver changes.
- Move "nvdimm/pmem: Fix leak on dax_add_host() failure" outside of this
  series.

Changes since v3:
- Fix a leak on dax_add_host() failure in nvdimm/pmem.
- Split the series into a bissectable sequence of changes.
- Ensure that device-dax use-cases still works on data cache
  aliasing architectures.

Changes since v2:
- Move DAX supported runtime check to alloc_dax(),
- Modify DM to handle alloc_dax() error as non-fatal,
- Remove all filesystem modifications, since the check is now done by
  alloc_dax(),
- rename "dcache" and "cache" to "cpu dcache" and "cpu cache" to
  eliminate confusion with VFS terminology.

Changes since v1:
- The order of the series was completely changed based on the
  feedback received on v1,
- cache_is_aliasing() is renamed to dcache_is_aliasing(),
- ARCH_HAS_CACHE_ALIASING_DYNAMIC is gone,
- dcache_is_aliasing() vs ARCH_HAS_CACHE_ALIASING relationship is
  simplified,
- the dax_is_supported() check was moved to its rightful place in all
  filesystems.

Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Vishal Verma <vishal.l.verma@intel.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Russell King <linux@armlinux.org.uk>
Cc: linux-arch@vger.kernel.org
Cc: linux-cxl@vger.kernel.org
Cc: linux-fsdevel@vger.kernel.org
Cc: linux-mm@kvack.org
Cc: linux-xfs@vger.kernel.org
Cc: dm-devel@lists.linux.dev
Cc: nvdimm@lists.linux.dev
Cc: linux-s390@vger.kernel.org

Mathieu Desnoyers (8):
  dax: alloc_dax() return ERR_PTR(-EOPNOTSUPP) for CONFIG_DAX=n
  nvdimm/pmem: Treat alloc_dax() -EOPNOTSUPP failure as non-fatal
  dm: Treat alloc_dax() -EOPNOTSUPP failure as non-fatal
  dcssblk: Handle alloc_dax() -EOPNOTSUPP failure
  virtio: Treat alloc_dax() -EOPNOTSUPP failure as non-fatal
  dax: Check for data cache aliasing at runtime
  Introduce cpu_dcache_is_aliasing() across all architectures
  dax: Fix incorrect list of data cache aliasing architectures

 arch/arc/Kconfig                    |  1 +
 arch/arc/include/asm/cachetype.h    |  9 +++++++++
 arch/arm/Kconfig                    |  1 +
 arch/arm/include/asm/cachetype.h    |  2 ++
 arch/csky/Kconfig                   |  1 +
 arch/csky/include/asm/cachetype.h   |  9 +++++++++
 arch/m68k/Kconfig                   |  1 +
 arch/m68k/include/asm/cachetype.h   |  9 +++++++++
 arch/mips/Kconfig                   |  1 +
 arch/mips/include/asm/cachetype.h   |  9 +++++++++
 arch/nios2/Kconfig                  |  1 +
 arch/nios2/include/asm/cachetype.h  | 10 ++++++++++
 arch/parisc/Kconfig                 |  1 +
 arch/parisc/include/asm/cachetype.h |  9 +++++++++
 arch/sh/Kconfig                     |  1 +
 arch/sh/include/asm/cachetype.h     |  9 +++++++++
 arch/sparc/Kconfig                  |  1 +
 arch/sparc/include/asm/cachetype.h  | 14 ++++++++++++++
 arch/xtensa/Kconfig                 |  1 +
 arch/xtensa/include/asm/cachetype.h | 10 ++++++++++
 drivers/dax/super.c                 | 14 ++++++++++++++
 drivers/md/dm.c                     | 17 +++++++++--------
 drivers/nvdimm/pmem.c               | 22 ++++++++++++----------
 drivers/s390/block/dcssblk.c        | 11 ++++++-----
 fs/Kconfig                          |  1 -
 fs/fuse/virtio_fs.c                 | 15 +++++++++++----
 include/linux/cacheinfo.h           |  6 ++++++
 include/linux/dax.h                 |  6 +-----
 mm/Kconfig                          |  6 ++++++
 29 files changed, 165 insertions(+), 33 deletions(-)
 create mode 100644 arch/arc/include/asm/cachetype.h
 create mode 100644 arch/csky/include/asm/cachetype.h
 create mode 100644 arch/m68k/include/asm/cachetype.h
 create mode 100644 arch/mips/include/asm/cachetype.h
 create mode 100644 arch/nios2/include/asm/cachetype.h
 create mode 100644 arch/parisc/include/asm/cachetype.h
 create mode 100644 arch/sh/include/asm/cachetype.h
 create mode 100644 arch/sparc/include/asm/cachetype.h
 create mode 100644 arch/xtensa/include/asm/cachetype.h