[Qemu-devel] [PATCH v4 00/11] Fix versatile_pci

[Peter Maydell]

This patch series fixes a number of serious bugs in our emulation of
the PCI controller found on VersatilePB and the early Realview boards:
 * our interrupt mapping was totally wrong
 * we weren't implementing the PCI memory windows
 * the I/O window wasn't mapped on VersatilePB
 * realview mapped things at the wrong addresses
 * we didn't implement the controller's registers at all
It also updates to some reasonable approximation to QOM best practice,
including subclassing pci_host rather than doing everything by hand.

I haven't implemented support for the SMAP registers (which control
how the controller converts accesses made by bus-mastering PCI
devices into system addresses). For the moment we rely on the fact
that Linux always maps things 1:1. (It wouldn't be too hard to add
SMAP support but it requires changing QEMU's pci code to allow a
controller to pass in a MemoryRegion* for DMA to use instead of
the system address space, so I prefer to leave that for another series.)

Patchset tested on both versatilepb and realview, using a set of
Linux kernel patches written by Arnd Bergmann:
http://lists.infradead.org/pipermail/linux-arm-kernel/2010-October/029040.html
which were in turn tested against real PB926 and PB1176 hardware.

This version of the patchset avoids breaking legacy Linux guests
by automatically detecting those broken kernels and switching back
to the old mapping. This works by looking at the values the kernel
writes to the PCI_INTERRUPT_LINE register in the config space, which
is effectively the interrupt number the kernel expects the device
to be using. If this doesn't match reality we use the broken mapping.
(Thanks to Michael S. Tsirkin for this suggestion.)

If the kernel ever writes something different to the old kernel
then QEMU will assume it is dealing with a working kernel and
will stop looking for broken values. This allows a new kernel to
put any values it likes in PCI_INTERRUPT_LINE, provided its init
code does an initial write of something other than 27 to let QEMU
know it can cope with the correct irq mapping.

Changes v3->v4:
 * rebase following recent pci changes
 * patch 7 minor tweaks as per mst review:
    added VPB_PCI prefix to enum names, drop specified values
    drop unnecessary casts of void*
Changes v2->v3:
 * patch 7: make broken guest autodetect a one-time thing rather than
   reversible, to allow more flexibility in how new kernels choose
   to use the PCI_INTERRUPT_LINE registers
Changes v1->v2:
 * autodetect broken guests rather than using user-set property
 * use pci_swizzle_map_fn
 * new patch: drop unnecessary vpb_pci_config_addr() function

Peter Maydell (11):
  versatile_pci: Fix hardcoded tabs
  versatile_pci: Expose PCI I/O region on Versatile PB
  versatile_pci: Update to realize and instance init functions
  versatile_pci: Change to subclassing TYPE_PCI_HOST_BRIDGE
  versatile_pci: Use separate PCI I/O space rather than system I/O space
  versatile_pci: Put the host bridge PCI device at slot 29
  versatile_pci: Implement the correct PCI IRQ mapping
  versatile_pci: Implement the PCI controller's control registers
  arm/realview: Fix mapping of PCI regions
  versatile_pci: Expose PCI memory space to system
  hw/versatile_pci: Drop unnecessary vpb_pci_config_addr()

 hw/arm/realview.c    |   22 +--
 hw/arm/versatilepb.c |   11 +-
 hw/versatile_pci.c   |  399 ++++++++++++++++++++++++++++++++++++++++++++------
 3 files changed, 371 insertions(+), 61 deletions(-)

-- 
1.7.9.5