This patchset fixes an error in how we were reporting the
architectural timers in the device tree for the virt board.
Old KVM host kernels (before 4.4) implemented these timers
as edge-triggered, which led us to decide to report them in
the DT as edge triggered. However in newer kernels this
was fixed and they're level triggered as they're architecturally
supposed to be.
Patch 4 fixes up the DT.
Patch 1 adds the 2.9 machine type so we can make the DT
change only apply to new virt machine types
Patches 2 and 3 are a cleanup, so that we can have
convenient access to the VirtMachineClass at the point
in the code where we want to look at it. We collapse the
VirtBoardInfo struct into the VirtMachineState, since
the former mostly exists because at the time there was
no VirtMachineState.
(Most of the diffstat is the result of s/vbi/vms/ on
the file to update variable names.)
Kernel behaviours as far as I know them:
* host v4.3 and earlier: implemented timers as weird
sort-of-edge-triggered semantics
* host v4.4 and later: timers are level triggered
vs
* guest pre-4.8: ignore the values in the DT, trust
whatever the GIC's reset values are
* guest 4.8 before rc6: write the DT values to the
GIC config registers
* guest newer than rc6 ignore the DT, and always write
'level triggered' to the GIC
This combination means that (assuming that post 4.8
guest kernels work at all on pre 4.4 host kernels) I
don't expect this to break anything. Testing would probably
be a good idea, though: I plan to do some more combinations
but kernel builds on my aarch64 system are slow so I
thought I'd send out the patchset before I went home today...
Peter Maydell (4):
hw/arm/virt: add 2.9 machine type
hw/arm/virt: Merge VirtBoardInfo and VirtMachineState
hw/arm/virt: Rename 'vbi' variables to 'vms'
hw/arm/virt: Don't incorrectly claim architectural timer to be
edge-triggered
include/hw/compat.h | 3 +
hw/arm/virt.c | 616 +++++++++++++++++++++++++++-------------------------
2 files changed, 323 insertions(+), 296 deletions(-)
--
2.7.4
