[v2,00/22] Add new DMA mapping operation for P2PDMA

Message

Logan Gunthorpe May 13, 2021, 10:31 p.m. UTC

Hi,

This patchset continues my work to to add P2PDMA support to the common
dma map operations. This allows for creating SGLs that have both P2PDMA
and regular pages which is a necessary step to allowing P2PDMA pages in
userspace.

The earlier RFC[1] and v1[2] postings generated a lot of great feedback.
This version adds a bunch more cleanup at the start of the series. I'll
probably look to split the earlier patches off and get them merged
indpendantly after a round of review with this series as this series
has gotten quite long.

I'm happy to do a few more passes if anyone has any further feedback
or better ideas.

This series is based on v5.13-rc1 and a git branch can be found here:

  https://github.com/sbates130272/linux-p2pmem/  p2pdma_map_ops_v2

Thanks,

Logan

[1] https://lore.kernel.org/linux-block/[e-mail]/
[2] https://lore.kernel.org/linux-block/[e-mail]/

Changes sine v1:
 * Rebased onto v5.13-rc1
 * Add some cleanup to the existing P2PDMA code to fix up some naming
   conventions and documentation as the code has evolved a bit since the
   names were chosen. (As suggested by John)
 * Add a patch that adds a warning if a host bridge is not in the whitelist
   (as suggested by Don)
 * Change to using dma_map_sgtable() instead of creating a new
   interface. For this, a couple of .map_sg implementations were changed
   to return full error codes. (as per Christoph)
 * Renamed the scatterlist functions to include the term "dma" to
   indicate that they apply to the DMA side of the sg. (per Jason)
 * Introduce ib_dma_pci_p2p_dma_supported() helper instead of open
   coding the check (per Jason)
 * Numerous minor adjustments and documentation fixes

Changes since the RFC:
 * Added comment and fixed up the pci_get_slot patch. (per Bjorn)
 * Fixed glaring sg_phys() double offset bug. (per Robin)
 * Created a new map operation (dma_map_sg_p2pdma()) with a new calling
   convention instead of modifying the calling convention of
   dma_map_sg(). (per Robin)
 * Integrated the two similar pci_p2pdma_dma_map_type() and
   pci_p2pdma_map_type() functions into one (per Ira)
 * Reworked some of the logic in the map_sg() implementations into
   helpers in the p2pdma code. (per Christoph)
 * Dropped a bunch of unnecessary symbol exports (per Christoph)
 * Expanded the code in dma_pci_p2pdma_supported() for clarity. (per
   Ira and Christoph)
 * Finished off using the new dma_map_sg_p2pdma() call in rdma_rw
   and removed the old pci_p2pdma_[un]map_sg(). (per Jason)

--

Logan Gunthorpe (22):
  PCI/P2PDMA: Rename upstream_bridge_distance() and rework documentation
  PCI/P2PDMA: Use a buffer on the stack for collecting the acs list
  PCI/P2PDMA: Cleanup type for return value of calc_map_type_and_dist()
  PCI/P2PDMA: Avoid pci_get_slot() which sleeps
  PCI/P2PDMA: Print a warning if the host bridge is not in the whitelist
  PCI/P2PDMA: Attempt to set map_type if it has not been set
  PCI/P2PDMA: Refactor pci_p2pdma_map_type() to take pagemap and device
  dma-mapping: Allow map_sg() ops to return negative error codes
  dma-direct: Return appropriate error code from dma_direct_map_sg()
  iommu: Return full error code from iommu_map_sg[_atomic]()
  dma-iommu: Return error code from iommu_dma_map_sg()
  lib/scatterlist: Add flag for indicating P2PDMA segments in an SGL
  PCI/P2PDMA: Make pci_p2pdma_map_type() non-static
  PCI/P2PDMA: Introduce helpers for dma_map_sg implementations
  dma-direct: Support PCI P2PDMA pages in dma-direct map_sg
  dma-mapping: Add flags to dma_map_ops to indicate PCI P2PDMA support
  iommu/dma: Support PCI P2PDMA pages in dma-iommu map_sg
  nvme-pci: Check DMA ops when indicating support for PCI P2PDMA
  nvme-pci: Convert to using dma_map_sgtable()
  RDMA/core: Introduce ib_dma_pci_p2p_dma_supported()
  RDMA/rw: use dma_map_sgtable()
  PCI/P2PDMA: Remove pci_p2pdma_[un]map_sg()

 drivers/infiniband/core/rw.c |  75 ++++------
 drivers/iommu/dma-iommu.c    |  86 +++++++++--
 drivers/iommu/iommu.c        |  15 +-
 drivers/nvme/host/core.c     |   3 +-
 drivers/nvme/host/nvme.h     |   2 +-
 drivers/nvme/host/pci.c      |  80 +++++-----
 drivers/nvme/target/rdma.c   |   2 +-
 drivers/pci/Kconfig          |   2 +-
 drivers/pci/p2pdma.c         | 273 +++++++++++++++++++----------------
 include/linux/dma-map-ops.h  |  18 ++-
 include/linux/dma-mapping.h  |  46 +++++-
 include/linux/iommu.h        |  22 +--
 include/linux/pci-p2pdma.h   |  81 ++++++++---
 include/linux/scatterlist.h  |  50 ++++++-
 include/rdma/ib_verbs.h      |  30 ++++
 kernel/dma/direct.c          |  44 +++++-
 kernel/dma/mapping.c         |  31 +++-
 17 files changed, 570 insertions(+), 290 deletions(-)


base-commit: 6efb943b8616ec53a5e444193dccf1af9ad627b5
--
2.20.1

Comments

Christoph Hellwig May 14, 2021, 2 p.m. UTC | #1

On Thu, May 13, 2021 at 04:31:41PM -0600, Logan Gunthorpe wrote:
>  17 files changed, 570 insertions(+), 290 deletions(-)

I'm a little worried about all this extra code for no new functionality
at all.

Logan Gunthorpe May 14, 2021, 4:30 p.m. UTC | #2

On 2021-05-14 8:00 a.m., Christoph Hellwig wrote:
> On Thu, May 13, 2021 at 04:31:41PM -0600, Logan Gunthorpe wrote:
>>  17 files changed, 570 insertions(+), 290 deletions(-)
> 
> I'm a little worried about all this extra code for no new functionality
> at all.

Yes, this series is really just prep work for allowing new
functionality. And a bunch of cleanup has been tacked onto it. It should
make adding P2PDMA in userspace a lot easier and I expect other P2PDMA
use cases will be better enabled by it.

A lot of people don't like the pci_p2pdma_map_sg() special case and
can't use it in their use case because it only accepts homogenous SGLs.
This series gets rid of the special case and allows it to be used more
generally.

Logan