Paolo Bonzini [Tue, 5 Jun 2012 22:04:55 +0000 (00:04 +0200)]
qcow2: always operate caches in writeback mode
Writethrough does not need special-casing anymore in the qcow2 caches.
The block layer adds flushes after every guest-initiated data write,
and these will also flush the qcow2 caches to the OS.
Paolo Bonzini [Tue, 5 Jun 2012 22:04:53 +0000 (00:04 +0200)]
block: always open drivers in writeback mode
Formats are entirely in charge of flushes for metadata writes. For
guest-initiated writes, a writethrough cache is faked in the block layer.
So we can always open in writeback mode.
Paolo Bonzini [Tue, 5 Jun 2012 22:04:50 +0000 (00:04 +0200)]
savevm: flush after saving vm state
Writing vm state uses bdrv_pwrite, so it will automatically get flushes
in writethrough mode. But doing a flush at the end in writeback mode
is probably a good idea anyway.
Paolo Bonzini [Tue, 5 Jun 2012 22:04:49 +0000 (00:04 +0200)]
block: flush in writethrough mode after writes
We want to make the formats handle their own flushes
autonomously, while keeping for guests the ability to use a writethrough
cache. Since formats will write metadata via bs->file, bdrv_co_do_writev
is the only place where we need to add a flush.
Paolo Bonzini [Wed, 6 Jun 2012 14:23:27 +0000 (16:23 +0200)]
qemu-iotests: start vms in qtest mode
This way, they will not execute any VM code at all. However, right now
the cancellation test is "relying" on being slowed down by TCG executing
BIOS code. So, change the timeouts.
Paolo Bonzini [Wed, 6 Jun 2012 14:23:26 +0000 (16:23 +0200)]
qemu-iotests: fill streaming test image with data
The TestStreamStop test case is racy; if the job completes before we can
cancel it, it fails. If we remove the sleep the job will be canceled
before it has even started, and the test succeeds but it is also not
testing anything interesting.
But if the image is left sparse, then the job has really nothing to do.
For qcow2 it will read one L2-table, for raw it will issue a bunch of
ioctls. This also falls under "not testing anything interesting", and
this may be happening right now (depending on the filesystem) since the
file protocol got an is_allocated method.
Filling the test image with data ensures that the test covers the
intended case. It also slows down the test, which will be particularly
important after the next patch.
Josh Durgin [Thu, 17 May 2012 20:42:29 +0000 (13:42 -0700)]
rbd: hook up cache options
Writeback caching was added in Ceph 0.46, and writethrough will be in
0.47. These are controlled by general config options, so there's no
need to check for librbd version.
Kevin Wolf [Fri, 11 May 2012 14:07:02 +0000 (16:07 +0200)]
qemu-img check -r for repairing images
The QED block driver already provides the functionality to not only
detect inconsistencies in images, but also fix them. However, this
functionality cannot be manually invoked with qemu-img, but the
check happens only automatically during bdrv_open().
This adds a -r switch to qemu-img check that allows manual invocation
of an image repair.
Max Filippov [Fri, 8 Jun 2012 09:50:34 +0000 (13:50 +0400)]
xtensa_lx60: add missing #include "blockdev.h"
This should fix the following build failure:
/home/buildbot/slave-public/block_mingw32/build/hw/xtensa_lx60.c: In function 'lx_init':
/home/buildbot/slave-public/block_mingw32/build/hw/xtensa_lx60.c:212: warning: implicit declaration of function 'drive_get'
/home/buildbot/slave-public/block_mingw32/build/hw/xtensa_lx60.c:212: warning: nested extern declaration of 'drive_get'
/home/buildbot/slave-public/block_mingw32/build/hw/xtensa_lx60.c:212: error: 'IF_PFLASH' undeclared (first use in this function)
/home/buildbot/slave-public/block_mingw32/build/hw/xtensa_lx60.c:212: error: (Each undeclared identifier is reported only once
/home/buildbot/slave-public/block_mingw32/build/hw/xtensa_lx60.c:212: error: for each function it appears in.)
/home/buildbot/slave-public/block_mingw32/build/hw/xtensa_lx60.c:216: error: dereferencing pointer to incomplete type
Paolo Bonzini [Tue, 8 May 2012 14:52:00 +0000 (16:52 +0200)]
stream: tweak usage of bdrv_co_is_allocated
is_allocated_base has complex semantics that are not really usable
outside streaming. Split the check in two parts, where the allocated
state for the top bs is moved to the caller. The resulting function
is more generally useful.
Kevin Wolf [Fri, 15 Jun 2012 11:43:18 +0000 (13:43 +0200)]
qcow2: Silence false warning
Some gcc versions seem not to be able to figure out that the switch
statement covers all possible values and that c is therefore always
initialised. Add a default branch for them.
Even though the xilinx tools do have C_ on all params by default, drop this
for consistency with all the other xilinx IP (I.E. param names are the xilinx
names without the C_ prefix)
Changed "txpingpong" prop to "tx-ping-pong". Same for rx. This is done to
make the property name exactly match what is output by the xilinx tools for
this IP.
The configurable property for this IP in the Xilinx tools is a boolean switch
"one-timer-only" that flicks this timer from being dual channel to single.
Updated QEMU to work the same way for better match with the IP core and its TRM.
The axidma irq orders are reversed in both the device model and the instantion.
Undid both reversal (for no net change). Also needs to be reversed for
consistency with Xilinx tools IRQ listing.
Anthony Liguori [Mon, 11 Jun 2012 17:15:51 +0000 (12:15 -0500)]
Merge remote-tracking branch 'afaerber-or/qom-cpu-3a' into staging
* afaerber-or/qom-cpu-3a: (27 commits)
target-s390x: Pass S390CPU to s390_cpu_restart()
s390-virtio: Let s390_cpu_addr2state() return S390CPU
s390-virtio: Use cpu_s390x_init() to obtain S390CPU
target-s390x: Let cpu_s390x_init() return S390CPU
xen_machine_pv: Use cpu_x86_init() to obtain X86CPU
arm_pic: Pass ARMCPU to arm_pic_init_cpu()
arm_boot: Pass ARMCPU to arm_load_kernel()
xilinx_zynq: Use cpu_arm_init() to obtain ARMCPU
pxa2xx_gpio: Store ARMCPU in PXA2xxGPIOInfo
pxa2xx_pic: Store ARMCPU in PXA2xxPICState
pxa2xx: Pass ARMCPU to pxa2xx_pic_init()
exynos4210: Use cpu_arm_init() to store ARMCPU
vexpress: Use cpu_arm_init() to obtain ARMCPU
realview: Use cpu_arm_init() to obtain ARMCPU
arm_boot: Pass ARMCPU to arm_boot_info::secondary_cpu_reset_hook()
arm_boot: Pass ARMCPU to arm_boot_info::write_secondary_boot()
versatilepb: Use cpu_arm_init() to obtain ARMCPU
musicpal: Use cpu_arm_init() to obtain ARMCPU
integratorcp: Use cpu_arm_init() to obtain ARMCPU
strongarm: Use cpu_arm_init() to store ARMCPU in StrongARMState
...
Anthony Liguori [Mon, 11 Jun 2012 17:15:51 +0000 (12:15 -0500)]
Merge remote-tracking branch 'afaerber-or/qom-next-1' into staging
* afaerber-or/qom-next-1:
target-i386: Use uint32 visitor for [x]level properties
qdev: Remove PropertyInfo range checking
qdev: Switch property accessors to fixed-width visitor interfaces
qdev: Use int32_t container for devfn property
qapi: Add String visitor coverage to serialization unit tests
qapi: String visitor, use %f representation for floats
qapi: Unit tests for visitor-based serialization
qapi: Add Visitor interfaces for uint*_t and int*_t
Jan Kiszka [Wed, 6 Jun 2012 14:28:42 +0000 (16:28 +0200)]
kvm: i8254: Fix conversion of in-kernel to userspace state
Due to a offset between the clock used to generate the in-kernel
count_load_time (CLOCK_MONOTONIC) and the clock used for processing this
in userspace (vm_clock), reading back the output of PIT channel 2 via
port 0x61 was broken. One use cases that suffered from it was the CPU
frequency calibration of SeaBIOS, which also affected IDE/AHCI timeouts.
This fixes it by calibrating the offset between both clocks on
kvm_pit_get and adjusting the kernel value before saving it in the
userspace state. As the calibration only works while the vm_clock is
running, we cache the in-kernel state across stopped phases.
Jan Beulich [Mon, 11 Jun 2012 09:52:27 +0000 (09:52 +0000)]
qemu/xendisk: set maximum number of grants to be used
Legacy (non-pvops) gntdev drivers may require this to be done when the
number of grants intended to be used simultaneously exceeds a certain
driver specific default limit.
Jim Meyering [Thu, 10 May 2012 16:19:47 +0000 (18:19 +0200)]
kvm/apic: correct short memset
kvm_put_apic_state's attempt to clear *kapic before setting its
bits cleared sizeof(void*) bytes (no more than 8) rather than the
intended 1024 (KVM_APIC_REG_SIZE) bytes. Spotted by coverity.
Max Filippov [Sun, 27 May 2012 23:18:33 +0000 (03:18 +0400)]
target-xtensa: fix CCOUNT for conditional branches
Taken conditional branches fail to update CCOUNT register because
accumulated ccount_delta is reset during translation of non-taken
branch. To fix it only update CCOUNT once per conditional branch
instruction translation.
This fixes guest linux freeze on LTP waitpid06 test.
Max Filippov [Sun, 27 May 2012 16:21:08 +0000 (20:21 +0400)]
exec: fix TB invalidation after breakpoint insertion/deletion
tb_invalidate_phys_addr has to be called with the exact physical address of
the breakpoint we add/remove, not just the page's base address.
Otherwise we easily fail to flush the right TB.
This breakage was introduced by the commit f3705d5329 "memory: make
phys_page_find() return an unadjusted".
This appeared to work for some guest architectures because their
cpu_get_phys_page_debug implementation returns full translated physical
address, not just the base of the TARGET_PAGE_SIZE-sized page.
Max Filippov [Sun, 27 May 2012 14:34:53 +0000 (18:34 +0400)]
target-xtensa: control page table lookup explicitly
Hardware pagetable walking may not be nested. Stop guessing and pass
explicit flag to the get_physical_addr_mmu function that controls page
table lookup.
Michael Tokarev [Sat, 2 Jun 2012 19:43:33 +0000 (23:43 +0400)]
do not include <libutil.h> needlessly or if it doesn't exist
<libutil.h> and <util.h> on *BSD (some have one, some another)
were #included just for openpty() declaration. The only file
where this function is actually used is qemu-char.c.
In vl.c and net/tap-bsd.c, none of functions declared in libutil.h
(login logout logwtmp timdomain openpty forkpty uu_lock realhostname
fparseln and a few others depending on version) are used.
Initially the code which is currently in qemu-char.c was in vl.c,
it has been removed into separate file in commit 0e82f34d077dc2542
Fri Oct 31 18:44:40 2008, but the #includes were left in vl.c.
So with vl.c, we just remove includes - libutil.h, util.h and
pty.h (which declares only openpty() and forkpty()) from there.
The code in net/tap-bsd.c, which come from net/tap.c, had this
Note this commit not only moved stuff out of net.c to net/tap.c,
but also rewrote large portions of the tap code, and added these
completely unnecessary #includes -- as usual, I question why such
a misleading commit messages are allowed.
Again, no functions defined in libutil.h or util.h on *BSD are
used by neither net/tap.c nor net/tap-bsd.c. Removing them.
And finally, the only real user for these #includes, qemu-char.c,
which actually uses openpty(). There, the #ifdef logic is wrong.
A GLIBC-based system has <pty.h>, even if it is a variant of *BSD.
So __GLIBC__ should be checked first, and instead of trying to
include <libutil.h> or <util.h>, we include <pty.h>. If it is not
GLIBC-based, we check for variations between <*util.h> as before.
This patch fixes build of qemu 1.1 on Debian/kFreebsd (well, one
of the two problems): it is a distribution with a FreeBSD kernel,
so it #defines at least __FreeBSD_kernel__, but since it is based
on GLIBC, it has <pty.h>, but current version does not have neither
<util.h> nor <libutil.h>, which the code tries to include 3 times
but uses only once.
Paolo Bonzini [Thu, 12 Apr 2012 13:32:26 +0000 (15:32 +0200)]
qdev: Remove PropertyInfo range checking
Range checking in PropertyInfo is now used only for pci_devfn
properties and some error reporting. Remove all code that implements
it in the various property types, and the now unused fields.
Signed-off-by: Paolo Bonzini <[email protected]>
[AF: Fix blocksize min/max for 32-bit hosts by using const int64_t.] Signed-off-by: Andreas Färber <[email protected]>
Michael Roth [Wed, 22 Feb 2012 18:26:37 +0000 (12:26 -0600)]
qdev: Switch property accessors to fixed-width visitor interfaces
This introduces {get,set}_uint{8,16,32,64}() functions for the
respective qdev types.
TADDR and VLAN are switched to explicit int64, BLOCKSIZE to uint16.
Michael Roth [Sun, 4 Mar 2012 19:38:27 +0000 (13:38 -0600)]
qdev: Use int32_t container for devfn property
Valid range for devfn is -1 to 255 (-1 for automatic assignment). We do
not currently validate this due to devfn being stored as a uint32_t.
This can lead to segfaults and other strange behavior.
We could technically just cast it to int32_t to implement the checking,
but this will not work for visitor-based setting where we may do additional
bounds-checking based on target container type, which is int32_t for this
case.
projects / qemu.git / log