[Qemu-devel] [PATCH 00/17] vfio: quirks & tracing refactoring

[Alex Williamson]

This series aims to do a few things:

 - Improve tracing support in vfio, especially the ability to isolate
   tracing to various features, like INTx, MSI, MSI-X, quirks,
   specific quirks, etc.  There's more to do here, but this is an
   improvement.

 - Split quirks out of the core vfio-pci code.  They're too large,
   they need to be separated.

 - Make quirks more maintainable and extensible.  In the past, I tried
   to make a common quirk interface but the commonality was pushed too
   high resulting in a shared data structure that didn't make sense
   for most of the users.  This version still provides sufficient
   commonality for shared teardown, but doesn't impose a data struct
   on memory region ops.

 - Make better use of MemoryRegions.  We can use more than a single
   MemoryRegion for each quirk, simplifying the ops callbacks to
   handle the behavior of only a single register.  For instance data
   and window registers can make use of two MemoryRegions.

 - Eliminate target page size dependencies in quirks.  Some of these
   must have been to work around bugs in previous generations of the
   memory API, some were just for convenience.  We don't need to
   setup quirks on target page boundaries, that's the memory API's
   problem.  We don't need to save bits and assume page alignment in
   the quirks, we're better off using structure specific to our quirk
   ops and trimming out unused features.

 - Eliminate build-time debugging and tunables.  Interrupt setup is
   slow path, so we're not saving anything by requiring a rebuild
   to not use the KVM interrupt path.  Make these all experimental
   (aka unsupported) options that are available for debugging and
   tracing.  Also fix the polarity of x-mmap and rename to x-no-mmap.
   Backend drivers shouldn't need to expose this unless they want the
   tracing ability and I've seen users specify it because mmap sounds
   like something they'd want.

Beyond the standard testing I've done, using a variety of devices and
configurations that rely on the quirks, I've received testing feed
back from several users on the vfio-users mailing list who have
reported no regressions.  Comments and review welcome.  Thanks,

Alex

PS - this series applies on top of 2.4 + "vfio/pci: Cleanup RTL8168
     quirk and tracing" + "vfio/pci: Cleanup vfio_early_setup_msix()
     error path".  Both of these are in my local next branch but
     haven't been pushed for 2.5 yet.

---

Alex Williamson (17):
      vfio/pci: Rename INTx functions for easier tracing
      vfio/pci: Rename MSI/X functions for easier tracing
      vfio/pci: Make interrupt bypass runtime configurable
      vfio: Change polarity of our no-mmap option
      vfio/pci: Extract PCI structures to a separate header
      vfio/pci: Split quirks to a separate file
      vfio/pci: Cleanup ROM blacklist quirk
      vfio/pci: Foundation for new quirk structure
      vfio/pci: Cleanup ATI 0x3c3 quirk
      vfio/pci: Cleanup Nvidia 0x3d0 quirk
      vfio/pci: Rework RTL8168 quirk
      vfio/pci: Config window quirks
      vfio/pci: Config mirror quirk
      vfio/pci: Remove old config window and mirror quirks
      vfio/pci: Move AMD device specific reset to quirks
      vfio/pci: Cache vendor and device ID
      vfio/pci: Add emulated PCI IDs


 hw/vfio/Makefile.objs         |    2 
 hw/vfio/common.c              |    2 
 hw/vfio/pci-quirks.c          | 1204 ++++++++++++++++++++++++++++++++++++
 hw/vfio/pci.c                 | 1385 ++++-------------------------------------
 hw/vfio/pci.h                 |  159 +++++
 hw/vfio/platform.c            |    2 
 include/hw/vfio/vfio-common.h |    7 
 trace-events                  |   85 +--
 8 files changed, 1546 insertions(+), 1300 deletions(-)
 create mode 100644 hw/vfio/pci-quirks.c
 create mode 100644 hw/vfio/pci.h