Peter Maydell [Wed, 26 Feb 2014 17:20:04 +0000 (17:20 +0000)]
target-arm: Implement AArch64 TTBR*
Implement the AArch64 TTBR* registers. For v7 these were already 64 bits
to handle LPAE, but implemented as two separate uint32_t fields.
Combine them into a single uint64_t which can be used for all purposes.
Since this requires touching every use, take the opportunity to rename
the field to the architectural name.
Peter Maydell [Wed, 26 Feb 2014 17:20:04 +0000 (17:20 +0000)]
target-arm: Implement AArch64 TCR_EL1
Implement the AArch64 TCR_EL1, which is the 64 bit view of
the AArch32 TTBCR. (The uses of the bits in the register are
completely different, but in any given situation the CPU will
always interpret them one way or the other. In fact for QEMU EL1
is always 64 bit, but we share the state field because this
is the correct mapping to permit a future implementation of EL2.)
We also make the AArch64 view the 'master' as far as migration
and reset is concerned.
Implement the AArch64 memory attribute registers. Since QEMU doesn't
model caches it does not need to care about memory attributes at all,
and we can simply make these read-as-written.
We did not previously implement the AArch32 versions of the MAIR
registers, which went unnoticed because of the overbroad TLB_LOCKDOWN
reginfo definition; provide them now to keep the 64<->32 register
relationship clear.
We already provided AMAIR registers for 32 bit as simple RAZ/WI;
extend that to provide a 64 bit RAZ/WI AMAIR_EL1.
Implement all the AArch64 cache invalidate and clean ops
(which are all NOPs since QEMU doesn't emulate the cache).
The only remaining unimplemented cache op is DC ZVA.
Peter Maydell [Wed, 26 Feb 2014 17:20:02 +0000 (17:20 +0000)]
target-arm: Implement AArch64 MIDR_EL1
Implement the AArch64 view of the MIDR system register
(for AArch64 it is a simple constant, unlike the complicated
mess that TI925 imposes on the 32-bit view).
Peter Maydell [Wed, 26 Feb 2014 17:20:01 +0000 (17:20 +0000)]
target-arm: A64: Make cache ID registers visible to AArch64
Make the cache ID system registers (CLIDR, CSSELR, CCSIDR, CTR)
visible to AArch64. These are mostly simple 64-bit extensions of the
existing 32 bit system registers and so can share reginfo definitions.
CTR needs to have a split definition, but we can clean up the
temporary user-mode implementation in favour of using the CPU-specified
reset value, and implement the system-mode-required semantics of
restricting its EL0 accessibility if SCTLR.UCT is not set.
Peter Maydell [Wed, 26 Feb 2014 17:20:01 +0000 (17:20 +0000)]
target-arm: Fix raw read and write functions on AArch64 registers
The raw read and write functions were using the ARM_CP_64BIT flag in
ri->type to determine whether to treat the register's state field as
uint32_t or uint64_t; however AArch64 register info structs don't use
that flag. Abstract out the "how big is the field?" test into a
function and fix it to work for AArch64 registers. For this to work
we must ensure that the reginfo structs put into the hashtable have
the correct state field for their use, not the placeholder STATE_BOTH.
Christoffer Dall [Wed, 26 Feb 2014 17:20:01 +0000 (17:20 +0000)]
hw: arm_gic_kvm: Add KVM VGIC save/restore logic
Save and restore the ARM KVM VGIC state from the kernel. We rely on
QEMU to marshal the GICState data structure and therefore simply
synchronize the kernel state with the QEMU emulated state in both
directions.
We take some care on the restore path to check the VGIC has been
configured with enough IRQs and CPU interfaces that we can properly
restore the state, and for separate set/clear registers we first fully
clear the registers and then set the required bits.
Christoffer Dall [Wed, 26 Feb 2014 17:20:00 +0000 (17:20 +0000)]
arm: vgic device control api support
Support creating the ARM vgic device through the device control API and
setting the base address for the distributor and cpu interfaces in KVM
VMs using this API.
Because the older KVM_CREATE_IRQCHIP interface needs the irq chip to be
created prior to creating the VCPUs, we first test if we can use the
device control API in kvm_arch_irqchip_create (using the test flag from
the device control API). If we cannot, it means we have to fall back to
KVM_CREATE_IRQCHIP and use the older ioctl at this point in time. If
however, we can use the device control API, we don't do anything and
wait until the arm_gic_kvm driver initializes and let that use the
device control API.
Christoffer Dall [Wed, 26 Feb 2014 17:19:59 +0000 (17:19 +0000)]
hw/intc/arm_gic: Fix GIC_SET_LEVEL
The GIC_SET_LEVEL macro unfortunately overwrote the entire level
bitmask instead of just or'ing on the necessary bits, causing active
level PPIs on a core to clear PPIs on other cores.
In ARMv5 level 2 page table descriptors, each 4K or 64K page is split into
four subpages, each of which can have different access permission settings,
which are specified by four two-bit fields in the l2 descriptor. A
long-standing cut-and-paste error meant we were using the wrong bits in
the virtual address to select the access-permission field for 4K pages.
The error has presumably not been noticed before because most guests don't
make use of the ability to set the access permissions differently for
each 1K subpage: if the guest gives the whole page the same access
permissions it doesn't matter which of the 4 AP fields we select.
(The whole issue is irrelevant for ARMv7 CPUs anyway because subpages
aren't supported there.)
The ethernet device in the musicpal only has two tx queues,
but we modelled it with four CTDP registers, presumably a
cut and paste from the rx queue registers. Since the tx_queue[]
array is only 2 entries long this allowed a guest to overrun
this buffer. Remove the nonexistent registers.
Peter Maydell [Wed, 26 Feb 2014 17:19:58 +0000 (17:19 +0000)]
hw/intc/exynos4210_combiner: Don't overrun output_irq array in init
The Exynos4210 combiner has IIC_NIRQ inputs and IIC_NGRP outputs;
use the correct constant in the loop initializing our output
sysbus IRQs so that we don't overrun the output_irq[] array.
Peter Maydell [Wed, 26 Feb 2014 17:19:58 +0000 (17:19 +0000)]
hw/timer/arm_timer: Avoid array overrun for bad addresses
The integrator's timer read/write functions log an error for
bad addresses in guest accesses, but were falling through and
using an out of bounds array index rather than returning early.
Fix this.
Add a cast to avoid an unintended sign extension that
would mean we returned 0xffffffff in the high 32 bits
for an IA0 read if bit 31 in the MAC address was 1.
(This is harmless since we'll only be doing 4 byte
reads, but it could be confusing, so best avoided.)
Peter Maydell [Wed, 26 Feb 2014 17:19:57 +0000 (17:19 +0000)]
hw/misc/arm_sysctl: Fix bad boundary check on mb clock accesses
Fix incorrect use of sizeof() rather than ARRAY_SIZE() to guard
accesses into the mb_clock[] array, which was allowing a malicious
guest to overwrite the end of the array.
Paolo Bonzini [Tue, 25 Feb 2014 16:36:55 +0000 (17:36 +0100)]
modules: do not include gmodule-2.0 in static builds
gmodule-2.0's pkg-config files include -Wl,--export-dynamic, which breaks
static builds. It is a glib bug, but we need to support --static builds for
the linux-user targets, and in the end all that is needed to fix this is:
* outlaw --enable-modules --static, which makes little sense anyway
* only include gmodule-2.0's cflags and ldflags if --enable-modules is
specified on the command line.
Peter Maydell [Tue, 25 Feb 2014 12:30:01 +0000 (12:30 +0000)]
Merge remote-tracking branch 'remotes/mdroth/qga-pull-2014-02-24' into staging
* remotes/mdroth/qga-pull-2014-02-24:
qemu-ga: isa-serial support on Windows
qga: Fix memory allocation pasto
qga: Don't require 'time' argument in guest-set-time command
qga: vss-win32: Fix interference with snapshot deletion by other VSS request
qga: vss-win32: Fix interference with snapshot creation by other VSS requesters
qga: vss-win32: Use NULL as an invalid pointer for OpenEvent and CreateEvent
Peter Maydell [Tue, 25 Feb 2014 11:54:40 +0000 (11:54 +0000)]
Merge remote-tracking branch 'remotes/xtensa/tags/20140224-xtensa' into staging
Xtensa fixes and improvements queue 2014-02-24:
- add support for ML605 and KC705 FPGA boards;
- flush opencores_eth queue when new RX descriptor is available;
- add basic checks to cache opcodes;
- make core configuration available to tests;
- implement HW config ID special registers.
# gpg: Signature made Mon 24 Feb 2014 00:52:42 GMT using RSA key ID F83FA044
# gpg: Good signature from "Max Filippov <[email protected]>"
# gpg: aka "Max Filippov <[email protected]>"
* remotes/xtensa/tags/20140224-xtensa:
target-xtensa: provide HW confg ID registers
target-xtensa: refactor standard core configuration
target-xtensa: add basic tests for cache opcodes
target-xtensa: allow using core configuration in tests
target-xtensa: add overridable test_init macro
target-xtensa: add basic checks to icache opcodes
target-xtensa: add basic checks to dcache opcodes
target-xtensa: add RRRI4 opcode format fields
opencores_eth: flush queue whenever can_receive can go from false to true
hw/xtensa: add support for ML605 and KC705 FPGA board
Peter Maydell [Tue, 25 Feb 2014 10:50:11 +0000 (10:50 +0000)]
Merge remote-tracking branch 'remotes/kevin/tags/for-upstream' into staging
Block patches
# gpg: Signature made Fri 21 Feb 2014 21:42:24 GMT using RSA key ID C88F2FD6
# gpg: Good signature from "Kevin Wolf <[email protected]>"
* remotes/kevin/tags/for-upstream: (54 commits)
iotests: Mixed quorum child device specifications
quorum: Simplify quorum_open()
quorum: Add unit test.
quorum: Add quorum_open() and quorum_close().
quorum: Implement recursive .bdrv_recurse_is_first_non_filter in quorum.
quorum: Add quorum_co_flush().
quorum: Add quorum_invalidate_cache().
quorum: Add quorum_getlength().
quorum: Add quorum mechanism.
quorum: Add quorum_aio_readv.
blkverify: Extract qemu_iovec_clone() and qemu_iovec_compare() from blkverify.
quorum: Add quorum_aio_writev and its dependencies.
quorum: Create BDRVQuorumState and BlkDriver and do init.
quorum: Create quorum.c, add QuorumChildRequest and QuorumAIOCB.
check-qdict: Test termination of qdict_array_split()
check-qdict: Adjust test for qdict_array_split()
qdict: Extract non-QDicts in qdict_array_split()
qemu-config: Sections must consist of keys
qemu-iotests: Check qemu-img command line parsing
qemu-img: Allow -o help with incomplete argument list
...
Peter Maydell [Mon, 24 Feb 2014 15:37:59 +0000 (15:37 +0000)]
Merge remote-tracking branch 'remotes/bonzini/configure' into staging
* remotes/bonzini/configure:
build: softmmu targets do not have a "main.o" file
configure: Disable libtool if -fPIE does not work with it (bug #1257099)
block: convert block drivers linked with libs to modules
Makefile: introduce common-obj-m and block-obj-m for DSO
Makefile: install modules with "make install"
module: implement module loading
rules.mak: introduce DSO rules
darwin: do not use -mdynamic-no-pic
block: use per-object cflags and libs
rules.mak: allow per object cflags and libs
rules.mak: fix $(obj) to a real relative path
util: Split out exec_dir from os_find_datadir
Peter Maydell [Thu, 20 Feb 2014 15:10:16 +0000 (15:10 +0000)]
configure: check that C++ compiler actually works
Check that the C++ compiler works with the C compiler; if it
does not, then don't pass CXX to the build process. This
fixes a regression where QEMU was no longer building if the
build environment didn't have a C++ compiler (introduced
in commit 3144f78b, which incorrectly assumed that rules.mak
would only see a non-empty $(CXX) if configure had actually
found a working C++ compiler).
Max Filippov [Sat, 15 Feb 2014 16:58:47 +0000 (20:58 +0400)]
target-xtensa: refactor standard core configuration
Coalesce all standard configuration sections into single
DEFAULT_SECTIONS macro for all cores. This allows to add new features in
a single place: overlay_tool.h
Max Filippov [Fri, 7 Feb 2014 11:57:22 +0000 (15:57 +0400)]
target-xtensa: add basic checks to icache opcodes
Check privilege level for privileged instructions (IHU, III, IIU and IPFL
are privileged), memory accessibility for instructions that reference memory
(IH* and IPFL) and windowed register validity for all instruction cache
instructions.
Max Filippov [Fri, 7 Feb 2014 11:57:22 +0000 (15:57 +0400)]
target-xtensa: add basic checks to dcache opcodes
Check privilege level for privileged instructions (DHI, DHU, DII, DIU, DIWB,
DIWBI, DPFL are privileged), memory accessibility for instructions that
reference memory (all DH* and DPFL) and windowed register validity for all
data cache instructions.
Max Filippov [Mon, 3 Feb 2014 03:57:55 +0000 (07:57 +0400)]
opencores_eth: flush queue whenever can_receive can go from false to true
The following registers control whether MAC can receive frames:
- MODER.RXEN bit that enables/disables receiver;
- TX_BD_NUM register that specifies number of RX descriptors.
Notify QEMU networking core when the MAC is ready to receive frames.
Discard frame and raise BUSY interrupt when the frame arrives but the
current RX descriptor is not empty.
qmp_guest_file_seek() allocates memory for a GuestFileRead object
instead of the GuestFileSeek object it actually uses. Harmless,
because the GuestFileRead is slightly larger.
Michal Privoznik [Fri, 31 Jan 2014 10:29:51 +0000 (11:29 +0100)]
qga: Don't require 'time' argument in guest-set-time command
As the description to the guest-set-time states, the command is
there to ease time synchronization after resume. If guest was
suspended for longer period of time, its system time can go off
so badly, that even NTP refuses to set it. That's why the command
was invented: to give users chance to set the time (not
necessarily 100% correct). However, there's is no real need for
us to require users to pass an arbitrary time. Especially if we
can read the correct value from RTC (boiling down to reading
host's time). Hence this commit enables logic:
Tomoki Sekiyama [Mon, 13 Jan 2014 17:25:39 +0000 (12:25 -0500)]
qga: vss-win32: Fix interference with snapshot deletion by other VSS request
When a VSS requester such as vshadow.exe or diskshadow.exe requests to
delete snapshots, qemu-ga VSS provider's DeleteSnapshots() is also called
and returns E_NOTIMPL, that makes the deletion fail.
To avoid this issue, return S_OK and set values that represent no snapshots
are deleted by qemu-ga VSS provider.
Tomoki Sekiyama [Mon, 13 Jan 2014 17:25:29 +0000 (12:25 -0500)]
qga: vss-win32: Fix interference with snapshot creation by other VSS requesters
When a VSS requester such as vshadow.exe or diskshadow.exe requests to
create disk snapshots, Windows may choose qemu-ga VSS provider if it is
only provider registered on the system. However, because it provides only a
function to freeze the filesystem, the snapshotting fails.
This patch adds a check into CQGAVssProvider::IsVolumeSupported() to reject
the request from other VSS requesters, so that the other provider is chosen.
The check of requester is done by confirming event channels between
qemu-ga's requester and provider established. To ensure that the events are
initialized when CQGAVssProvider::IsVolumeSupported() is called, it moves
the initialization earlier.
Tomoki Sekiyama [Mon, 13 Jan 2014 17:25:23 +0000 (12:25 -0500)]
qga: vss-win32: Use NULL as an invalid pointer for OpenEvent and CreateEvent
OpenEvent and CreateEvent WinAPI return NULL when failed to open/create
events handles, instead of INVALID_HANDLE_VALUE (although their return
types are HANDLE).
This replaces INVALID_HANDLE_VALUE related to event handles with NULL.
Max Reitz [Fri, 21 Feb 2014 21:30:37 +0000 (22:30 +0100)]
quorum: Simplify quorum_open()
Although it may not look like it, this patch simplifies quorum_open().
qdict_array_split() is now able to return QLists with different objects
than only QDicts, therefore it will now do all the work and
quorum_open() does not have to handle reference strings by itself.
This allows mixing full option dicts and reference strings for
specifying the child block devices of quorum; furthermore, it improves
handling of malformed specifications.
Benoît Canet [Fri, 21 Feb 2014 21:21:15 +0000 (22:21 +0100)]
quorum: Add quorum mechanism.
This patchset enables the core of the quorum mechanism.
The num_children reads are compared to get the majority version and if this
version exists more than threshold times the guest won't see the error at all.
If a block is corrupted or if an error occurs during an IO or if the quorum
cannot be established QMP events are used to report to the management.
Use gnutls's SHA-256 to compare versions.
--enable-quorum must be used to enable the feature.
Benoît Canet [Fri, 21 Feb 2014 21:21:10 +0000 (22:21 +0100)]
quorum: Create quorum.c, add QuorumChildRequest and QuorumAIOCB.
Quorum is a block filter mirroring writes to num_children children.
For reads quorum reads each children and does a vote.
If more than vote_threshold versions are identical the quorum is reached and
this winning version is returned to the guest. So quorum prevents bit corruption.
For high availability purpose minority errors are reported via QMP but the guest
does not see them.
This patch creates the driver C source file and introduces the structures that
will be used in asynchronous reads and writes.
Max Reitz [Fri, 21 Feb 2014 18:11:40 +0000 (19:11 +0100)]
qdict: Extract non-QDicts in qdict_array_split()
Currently, qdict_array_split() only splits off entries with a key prefix
of "%u.", packing them into a new QDict. This patch makes it support
entries with the plain key "%u" as well, directly putting them into the
new QList without creating a QDict.
If there is both an entry with a key of "%u" and other entries with keys
prefixed "%u." (for the same index), the function simply terminates.
To do this, this patch also adds a static function which tests whether a
given QDict contains any keys with the given prefix. This is used to test
whether entries with a key prefixed "%u." do exist in the source QDict
without modifying it.
Max Reitz [Fri, 21 Feb 2014 18:11:39 +0000 (19:11 +0100)]
qemu-config: Sections must consist of keys
In config_parse_qdict_section(), the QList returned by
qdict_array_split() is assumed to only contain QDicts. Currently, this
is true but it may (and will) change in the future. Therefore, check
whether the assumption actually holds.
Kevin Wolf [Fri, 21 Feb 2014 15:24:07 +0000 (16:24 +0100)]
qemu-img: Allow -o help with incomplete argument list
This patch allows using 'qemu-img $subcmd -o help' for the create,
convert and amend subcommands, without specifying the previously
required filename arguments.
Note that it's still allowed and meaningful to specify a filename: An
invocation like 'qemu-img create -o help sheepdog:foo' will also display
options that are provided by the Sheepdog driver.
Kevin Wolf [Fri, 21 Feb 2014 15:24:06 +0000 (16:24 +0100)]
qemu-img amend: Support multiple -o options
Instead of ignoring all option values but the last one, multiple -o
options now have the same meaning as having a single option with all
settings in the order of their respective -o options.
Kevin Wolf [Fri, 21 Feb 2014 15:24:05 +0000 (16:24 +0100)]
qemu-img convert: Support multiple -o options
Instead of ignoring all option values but the last one, multiple -o
options now have the same meaning as having a single option with all
settings in the order of their respective -o options.
Kevin Wolf [Fri, 21 Feb 2014 15:24:03 +0000 (16:24 +0100)]
qemu-option: has_help_option() and is_valid_option_list()
has_help_option() checks if any help option ('help' or '?') occurs
anywhere in an option string, so that things like 'cluster_size=4k,help'
are recognised.
is_valid_option_list() ensures that the option list doesn't have options
with leading commas or trailing unescaped commas.
Paolo Bonzini [Mon, 17 Feb 2014 13:44:06 +0000 (14:44 +0100)]
block: do not abuse EMEDIUMTYPE
Returning "Wrong medium type" for an image that does not have a valid
header is a bit weird. Improve the error by mentioning what format
was trying to open it.
Paolo Bonzini [Mon, 17 Feb 2014 13:44:03 +0000 (14:44 +0100)]
vmdk: push vmdk_read_desc up to caller
Currently, we just try reading a VMDK file as both image and descriptor.
This makes it hard to choose which of the two attempts gave the best error.
We'll decide in advance if the file looks like an image or a descriptor,
and this patch is the first step to that end.
Paolo Bonzini [Mon, 17 Feb 2014 13:44:01 +0000 (14:44 +0100)]
vvfat: correctly propagate errors
Before:
$ ./qemu-io-old
qemu-io-old> open -r -o driver=vvfat,fat-type=24,dir=i386-softmmu
Valid FAT types are only 12, 16 and 32
qemu-io-old: can't open device (null): Could not open image: Invalid argument
After:
$ ./qemu-io
qemu-io> open -r -o driver=vvfat,fat-type=24,dir=i386-softmmu
qemu-io: can't open device (null): Valid FAT types are only 12, 16 and 32
Paolo Bonzini [Mon, 17 Feb 2014 13:43:54 +0000 (14:43 +0100)]
gluster: default scheme to gluster:// and host to localhost.
Currently, "gluster:///volname/img" and (using file. options)
"file.driver=gluster,file.filename=foo" will segfault. Also,
"//host/volname/img" will be rejected, but it is a valid URL
that should be accepted just fine with "file.driver=gluster".
Accept all of these, by inferring missing transport and host
as TCP and localhost respectively.
Paolo Bonzini [Mon, 17 Feb 2014 13:43:53 +0000 (14:43 +0100)]
iscsi: correctly propagate errors in iscsi_open
Before:
$ ./qemu-io-old
qemu-io-old> open -r -o file.driver=iscsi,file.filename=foo
Failed to parse URL : foo
qemu-io-old: can't open device (null): Could not open 'foo': Invalid argument
After:
$ ./qemu-io
qemu-io> open -r -o file.driver=iscsi,file.filename=foo
qemu-io: can't open device (null): Failed to parse URL : foo
projects / qemu.git / log