I've been accumulating a backlog of vfio-pci changes while we've been
getting the first implementation accepted. This series includes a
new algorithm for dealing with INTx that should avoid users needing
to know about experimental parameters, a re-try mechanism for mappings
that avoids common errors users might see trying to use vfio-pci, a
rework of MSIX setup and teardown that fixes a corner case for MSIX
being enabled but not used, a more robust device reset that re-inits
interrupts much like pci-assign, a missing INTx setup failure cleanup,
and addresses a few comments collected during initial review.
Comments welcome. If there are no changes or objections I'll send a
pull request for this next week. Thanks,
Alex
---
Alex Williamson (11):
vfio-pci: Cleanup on INTx setup failure
vfio-pci: Extend reset
vfio-pci: Remove setting of MSI qsize
vfio-pci: Use uintptr_t for void* cast
vfio-pci: Don't peak at msi_supported
vfio-pci: Roll the header into the .c file
vfio-pci: No spurious MSIs
vfio-pci: Rework MSIX setup/teardown
vfio-pci: Unmap and retry DMA mapping
vfio-pci: Re-order map/unmap
vfio-pci: Update slow path INTx algorithm
hw/vfio_pci.c | 395 +++++++++++++++++++++++++++++++++--------------------
hw/vfio_pci_int.h | 114 ---------------
2 files changed, 246 insertions(+), 263 deletions(-)
delete mode 100644 hw/vfio_pci_int.h
