[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Qemu-devel] [PATCH 00/28] Memory API for 1.6: fix I/O port endianness m
From: |
Paolo Bonzini |
Subject: |
[Qemu-devel] [PATCH 00/28] Memory API for 1.6: fix I/O port endianness mess |
Date: |
Mon, 22 Jul 2013 15:54:10 +0200 |
This series fixes the I/O port endianness problems that came when
port I/O was made to go through the normal dispatch path. Targets
that used isa_mmio to invoke cpu_inl and friends now tried to do
byte swapping once more than they previously did, causing garbage
to be written.
This series drops isa_mmio and replace it with an alias of the
root I/O memory region. After applying back the LITTLE_ENDIAN
mark for PortioLists, everything works as expected: the port memory
regions appear directly in the FlatView with the right endianness,
and MMIO is dispatched straight to them.
PReP cannot use this due to the non-contiguous map, so we just
make sure the prep_io_ops are marked DEVICE_NATIVE_ENDIAN; the swap
will happen when prep_io_ops redispatch to the device.
So we had a change that broke multiple ports, most of them
under-maintained. What better occasion to add a unit test? This
series does that, using the pc-testdev device as the harness.
The test was run before and after the I/O changes (in particular
on top of commit a014ed0, memory: accept mismatching sizes in
memory_region_access_valid, 2013-05-24), and passes there too.
This also found some unrelated problems caused by the new QOM cast in
commit 7588e2b (pci: Fold host_buses list into PCIHostState functionality,
2013-06-06).
I also smoke-tested prep, with the kernel that Herve provided, and
pseries. VGA works for both, which tests the PortioList path.
Alexey tested Linux (with virtio) on pseries.
Patches 1 to 15 fix all bugs and remove isa_mmio.
Patches 16 to 23 prepare the boards so that the testcase runs
successfully. This includes adding missing devices and unbreaking
other breakage.
Patch 24 adds the test case, patches 25-27 fix big-endian split/combine,
patch 28 adds a test case for that too.
If preferred, I can prepare a pull request that merges the tests
from an earlier branchpoint. This makes it easy to run the tests
on both the old and the new code.
Paolo
Alexey Kardashevskiy (1):
spapr_pci: remove indirection for I/O port access
Paolo Bonzini (27):
sh4: do not use isa_mmio
ppc_oldworld: do not use isa_mmio
ppc_newworld: do not use isa_mmio
prep: fix I/O port endianness
mips_jazz: do not use isa_mmio
mips_r4k: do not use isa_mmio
mips_malta: do not use isa_mmio
ppc440_bamboo: do not use isa_mmio
mipssim: do not use isa_mmio
mips_fulong2e: do not use isa_mmio
sparc64: remove indirection for I/O port access
ebus: do not use isa_mmio
isa_mmio: delete
Revert "ioport: remove LITTLE_ENDIAN mark for portio"
pc-testdev: support 8 and 16-bit accesses to 0xe0
pc-testdev: remove useless cpu_to_le64/le64_to_cpu
mips: degrade BIOS error to warning
sh4: unbreak r2d
sparc64: unbreak
default-configs: add test device to all machines supporting ISA
default-configs: add SuperIO to SH4
default-configs/ppc64: add all components of i82378 SuperIO chip used
by prep
qtest: add test for ISA I/O space endianness
memory: move functions around
memory: pass MemoryRegion to access_with_adjusted_size
memory: check memory region endianness, not target's
pc-testdev: add I/O port to test memory.c auto split/combine
default-configs/alpha-softmmu.mak | 1 +
default-configs/mips-softmmu.mak | 2 +-
default-configs/mips64-softmmu.mak | 2 +-
default-configs/mips64el-softmmu.mak | 2 +-
default-configs/mipsel-softmmu.mak | 2 +-
default-configs/ppc-softmmu.mak | 1 +
default-configs/ppc64-softmmu.mak | 7 +
default-configs/ppcemb-softmmu.mak | 1 +
default-configs/sh4-softmmu.mak | 9 +-
default-configs/sh4eb-softmmu.mak | 9 +-
default-configs/sparc64-softmmu.mak | 1 +
hw/isa/Makefile.objs | 1 -
hw/isa/isa_mmio.c | 81 ---------
hw/mips/gt64xxx_pci.c | 3 +-
hw/mips/mips_fulong2e.c | 3 +-
hw/mips/mips_jazz.c | 8 +-
hw/mips/mips_malta.c | 3 +-
hw/mips/mips_mipssim.c | 8 +-
hw/mips/mips_r4k.c | 6 +-
hw/misc/pc-testdev.c | 28 +++-
hw/pci-host/apb.c | 101 ++++-------
hw/pci-host/bonito.c | 25 ++-
hw/ppc/mac_newworld.c | 5 +-
hw/ppc/mac_oldworld.c | 5 +-
hw/ppc/ppc440_bamboo.c | 5 +-
hw/ppc/prep.c | 2 +-
hw/ppc/spapr_pci.c | 41 +----
hw/sh4/sh_pci.c | 42 +++--
hw/sparc64/sun4u.c | 6 +-
include/hw/isa/isa.h | 3 -
ioport.c | 1 +
memory.c | 101 +++++------
tests/Makefile | 14 +-
tests/endianness-test.c | 316 +++++++++++++++++++++++++++++++++++
34 files changed, 543 insertions(+), 302 deletions(-)
delete mode 100644 hw/isa/isa_mmio.c
create mode 100644 tests/endianness-test.c
--
1.8.1.4
- [Qemu-devel] [PATCH 00/28] Memory API for 1.6: fix I/O port endianness mess,
Paolo Bonzini <=
- [Qemu-devel] [PATCH 05/28] prep: fix I/O port endianness, Paolo Bonzini, 2013/07/22
- [Qemu-devel] [PATCH 06/28] mips_jazz: do not use isa_mmio, Paolo Bonzini, 2013/07/22
- [Qemu-devel] [PATCH 07/28] mips_r4k: do not use isa_mmio, Paolo Bonzini, 2013/07/22
- [Qemu-devel] [PATCH 09/28] ppc440_bamboo: do not use isa_mmio, Paolo Bonzini, 2013/07/22
- [Qemu-devel] [PATCH 08/28] mips_malta: do not use isa_mmio, Paolo Bonzini, 2013/07/22
- [Qemu-devel] [PATCH 10/28] mipssim: do not use isa_mmio, Paolo Bonzini, 2013/07/22
- [Qemu-devel] [PATCH 11/28] mips_fulong2e: do not use isa_mmio, Paolo Bonzini, 2013/07/22
- [Qemu-devel] [PATCH 12/28] sparc64: remove indirection for I/O port access, Paolo Bonzini, 2013/07/22
- [Qemu-devel] [PATCH 13/28] ebus: do not use isa_mmio, Paolo Bonzini, 2013/07/22
- [Qemu-devel] [PATCH 15/28] Revert "ioport: remove LITTLE_ENDIAN mark for portio", Paolo Bonzini, 2013/07/22