qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Qemu-devel] [PATCH v7 0/3] EEH Support for VFIO devices on sPAPR


From: Gavin Shan
Subject: [Qemu-devel] [PATCH v7 0/3] EEH Support for VFIO devices on sPAPR
Date: Tue, 27 May 2014 18:51:07 +1000

EEH Support for VFIO PCI Device

The series of patches intend to support EEH for VFIO PCI devices on sPAPR
platform. It requires corresponding host kernel support. Also, it was built
on top of Alexey's VFIO-for-sPAPR git repository.

QEMU:   git://github.com/aik/qemu.git (branch: vfio)
Kernel: 
http://linuxppc.10917.n7.nabble.com/PATCH-v7-0-3-EEH-Support-for-VFIO-PCI-Device-td82692.html

Here are some highlights and I don't have experience with QEMU. Any comments
should be appreciated.

* EEH RTAS calls captured by sPAPR platform firstly, and then routed to
  VFIO for handling if VFIO can. Otheriwse, it will be routed to host
  kernel for handling via dedicated VFIO container IOCTL command.
* PHB BUID is the identifier to one specific PHB, which is required information
  for EEH address mapping. In order to retrieve that from VFIO, the PCI
  bus struct has been extended to have additional field "sysdata" for that.
* The guest can't do error injection until holding valid token. Each guest
  is allowed to have one token at once. The RTAS calls managing token is
  simply emulated in QEMU. However, the error injection has to be routed
  to host kernel by dedicated syscall.

Changelog
=========
v3 -> v4:
        * Remove the error injection from the patchset. Mike or I will work
          on it later.
        * The IOCTL command is handled by VFIO PCI device instead of container.
        * "sysdata" in root bus to pass PHB BUID isn't needed.
        * Update the IOCTL argunemt according to kernel's changes.
v4 -> v6:
        * Header file update according to kerenl's change.
        * Split one ioctl command to 5.
        * Doesn't rely the kerenl to figure out RTAS return value. QEMU 
translate
          return value from ioctl() to RTAS return value insteadly.
v6 -> v7:
        * All ioctl commands go to VFIO container fd instead of PCI device fd.
        * QEMU figures out the PE address directly.
        * Rely on sPAPRVFIOPHBState, which is one-to-one mapping with IOMMU 
group.

Gavin Shan (3):
  headers: Sync with Linux header
  VFIO: Helper function to retrieve container fd
  sPAPR: EEH support for VFIO PCI device

 hw/misc/vfio.c              |  17 +++
 hw/ppc/spapr.c              |   1 +
 hw/ppc/spapr_pci.c          |   1 +
 hw/ppc/spapr_pci_vfio.c     | 332 ++++++++++++++++++++++++++++++++++++++++++++
 include/hw/misc/vfio.h      |   1 +
 include/hw/pci-host/spapr.h |   1 +
 include/hw/ppc/spapr.h      |   1 +
 linux-headers/linux/vfio.h  |  66 +++++++++
 8 files changed, 420 insertions(+)

-- 
1.8.3.2




reply via email to

[Prev in Thread] Current Thread [Next in Thread]