Michael Roth [Tue, 17 Apr 2012 00:52:17 +0000 (19:52 -0500)]
qemu-ga: improve recovery options for fsfreeze
guest-fsfreeze-thaw relies on state information obtained from
guest-fsfreeze-freeze to determine what filesystems to unfreeze.
This is unreliable due to the fact that that state does not account
for FIFREEZE being issued by other processes, or previous instances
of qemu-ga. This means in certain situations we cannot thaw
filesystems even with a responsive qemu-ga instance at our disposal.
This patch allows guest-fsfreeze-thaw to be issued unconditionally.
It also adds some additional logic to allow us to thaw filesystems
regardless of how many times the filesystem's "frozen" refcount has
been incremented by any guest processes.
Also, guest-fsfreeze-freeze now operates atomically: on success all
freezable filesystems are frozen, and on error all filesystems are
thawed. The ambiguous "GUEST_FSFREEZE_STATUS_ERROR" state is no
longer entered.
Anthony Liguori [Fri, 27 Apr 2012 17:00:06 +0000 (12:00 -0500)]
Merge remote-tracking branch 'qmp/queue/qmp' into staging
* qmp/queue/qmp:
qapi: fix qmp_balloon() conversion
qemu-iotests: add block-stream speed value test case
block: add 'speed' optional parameter to block-stream
block: change block-job-set-speed argument from 'value' to 'speed'
block: use Error mechanism instead of -errno for block_job_set_speed()
block: use Error mechanism instead of -errno for block_job_create()
Stefan Hajnoczi [Wed, 25 Apr 2012 15:51:04 +0000 (16:51 +0100)]
qemu-iotests: add block-stream speed value test case
Add tests to exercise the InvalidParameter 'speed' error code path, as
well as the regular success case for setting the speed. The
block-stream 'speed' parameter allows the speed limit of the job to be
applied immediately when the job starts instead of issuing a separate
block-job-set-speed command later. If the parameter has an invalid
value we expect to get an error and the job is not created.
It turns out that cancelling a block job is a common operation in these
test cases, let's extract a cancel_and_wait() function instead of
duplicating the QMP commands.
Stefan Hajnoczi [Wed, 25 Apr 2012 15:51:03 +0000 (16:51 +0100)]
block: add 'speed' optional parameter to block-stream
Allow streaming operations to be started with an initial speed limit.
This eliminates the window of time between starting streaming and
issuing block-job-set-speed. Users should use the new optional 'speed'
parameter instead so that speed limits are in effect immediately when
the job starts.
Stefan Hajnoczi [Wed, 25 Apr 2012 15:51:01 +0000 (16:51 +0100)]
block: use Error mechanism instead of -errno for block_job_set_speed()
There are at least two different errors that can occur in
block_job_set_speed(): the job might not support setting speeds or the
value might be invalid.
Use the Error mechanism to report the error where it occurs.
Stefan Hajnoczi [Wed, 25 Apr 2012 15:51:00 +0000 (16:51 +0100)]
block: use Error mechanism instead of -errno for block_job_create()
The block job API uses -errno return values internally and we convert
these to Error in the QMP functions. This is ugly because the Error
should be created at the point where we still have all the relevant
information. More importantly, it is hard to add new error cases to
this case since we quickly run out of -errno values without losing
information.
Go ahead and use Error directly and don't convert later.
Anthony Liguori [Thu, 26 Apr 2012 20:21:52 +0000 (15:21 -0500)]
Merge remote-tracking branch 'kraxel/usb.49' into staging
* kraxel/usb.49:
usb-uhci: update irq line on reset
usb: add serial number generator
usb-redir: Not finding an async urb id is not an error
usb-redir: Reset device address and speed on disconnect
usb-redir: An interface count of 0 is a valid value
usb-xhci: fix bit test
usb-xhci: Use PCI DMA helper functions
usb-host: fix zero-length packets
usb-host: don't dereference invalid iovecs
usb-storage: fix request canceling
usb-ehci: Ensure frindex writes leave a valid frindex value
usb-ehci: add missing usb_packet_init() call
usb-ehci: remove hack
Anthony Liguori [Thu, 26 Apr 2012 20:20:33 +0000 (15:20 -0500)]
Merge remote-tracking branch 'mst/tags/for_anthony' into staging
* mst/tags/for_anthony:
e1000: set E1000_ICR_INT_ASSERTED only for 8257x
e1000: link auto-negotiation emulation
e1000: introduce bit for debugging PHY emulation
e1000: introduce helpers to manipulate link status
e1000: PHY loopback mode support
e1000: conditionally raise irq at the end of MDI cycle
e1000: introduce bits of PHY control register
eepro100: Fix multicast regression
virtio: order index/descriptor reads
virtio: add missing mb() on enable notification
virtio: add missing mb() on notification
e1000: move reset function earlier in file
Anthony Liguori [Thu, 26 Apr 2012 20:18:06 +0000 (15:18 -0500)]
Merge remote-tracking branch 'afaerber/qom-cpu-x86-prop.v3' into staging
* afaerber/qom-cpu-x86-prop.v3:
target-i386: Introduce "tsc-frequency" property for X86CPU
target-i386: Prepare "vendor" property for X86CPU
target-i386: Introduce "xlevel" property for X86CPU
target-i386: Introduce "level" property for X86CPU
target-i386: Add property getter for CPU model-id
target-i386: Add property getter for CPU stepping
target-i386: Add property getter for CPU model
target-i386: Add property getter for CPU family
target-i386: Add "model-id" property to X86CPU
target-i386: Add "stepping" property to X86CPU
target-i386: Add "model" property to X86CPU
target-i386: Add "family" property to X86CPU
target-i386: Add range check for -cpu ,family=x
target-i386: Pass X86CPU to cpu_x86_register()
target-i386: Fix x86_cpuid_set_model_id()
timers: the rearm function should be able to handle delta = INT64_MAX
Fix win32_rearm_timer and mm_rearm_timer: they should be able to handle
INT64_MAX as a delta parameter without overflowing.
Also, the next deadline in ms should be calculated rounding down rather
than up (see unix_rearm_timer and dynticks_rearm_timer).
Finally ChangeTimerQueueTimer takes an unsigned long and timeSetEvent
takes an unsigned int as delta, so cast the ms delta to the appropriate
unsigned integer.
Stefan Weil [Fri, 13 Apr 2012 20:24:42 +0000 (22:24 +0200)]
sdl: Add QEMU mascot icon for use with SDL
This is a bitmap file (32x32x4) derived from the official QEMU mascot
(which was designed by Benoît Canet). I stripped the text from the SVG
to get a nearly square image and converted the result to BMP without
any manual optimization.
The bitmap is currently used by QEMU's SDL interface and replaces the
default X icon.
Anthony Liguori [Wed, 18 Apr 2012 22:33:15 +0000 (17:33 -0500)]
pc-sysfw: make sure to call qdev_init
We're not actually calling qdev_init for the pc-sysfw device. Since we create
the canonical path during realize, this was causing an assert to trigger when
attempting to read a link pointing to pc-sysfw.
This patch adds a function which creates unique serial numbers for usb
devices and puts it into use. Windows guests tend to become unhappy if
they find two identical usb devices in the system. Effects range from
non-functional devices (with yellow exclamation mark in device manager)
to BSODs. Handing out unique serial numbers to devices fixes this.
With this patch applied almost all emulated devices get a generated,
unique serial number. There are two exceptions:
* usb-storage devices will prefer a user-specified serial number
and will only get a generated number in case the serial property
is unset.
* usb-hid devices keep the fixed serial number "42" as it is used
to signal "remote wakeup actually works".
See commit 7b074a22dab4bdda9864b933f1bc811a3db42845
Hans de Goede [Sat, 31 Mar 2012 15:14:23 +0000 (17:14 +0200)]
usb-redir: Not finding an async urb id is not an error
We clear our pending async urb list on device disconnect and we may still
receive "packet complete" packets from our peer after this, which will then
refer to packet ids no longer in our list.
Hans de Goede [Sat, 31 Mar 2012 15:14:21 +0000 (17:14 +0200)]
usb-redir: An interface count of 0 is a valid value
An interface-count of 0 happens when a device is in unconfigured state when
it gets redirected. So we should not use 0 to detect not having received
interface info from our peer.
David Gibson [Wed, 4 Apr 2012 00:15:58 +0000 (10:15 +1000)]
usb-xhci: Use PCI DMA helper functions
Shortly before 1.0, we added helper functions / wrappers for doing PCI DMA
from individual devices. This makes what's going on clearer and means that
when we add IOMMU support somewhere in the future, only the general PCI
code will have to change, not every device that uses PCI DMA.
However, usb-xhci is not using these wrappers, despite being a PCI only
device. This patch remedies the situation, using the pci dma functions
instead of direct calls to cpu_physical_memory_{read,write}(). Likewise
address parameters for DMA are changed to dma_addr_t instead of
target_phys_addr_t.
usb-host optimizes away zero-length packets by not entering the
processing loop at all. Which isn't correct, we should submit a
zero-length urb to the host devicein that case. This patch makes
sure we run the processing loop at least once.
Little fix for usb packet handling on i/o cancelation. The
usb packet pointer (s->packet) is cleared at the wrong place:
The scsi request cancel handler does it. When a usb packet
is canceled the usb-storage emulation canceles the scsi request
if present. In most cases there is one, so usually s->packet
is cleared as needed even with the code sitting at the wrong
place.
If there is no scsi request in flight s->packet is not cleared
though. The usb-storage emulation will then try to complete an
usb packet which is not in flight any more and thereby trigger
an assert() in the usb core.
Fix this by clearing s->packet at the correct place, which is
the usb packet cancel header.
Hans de Goede [Tue, 3 Apr 2012 12:21:47 +0000 (14:21 +0200)]
usb-ehci: Ensure frindex writes leave a valid frindex value
frindex is a 14 bits counter, so bits 31-14 should always be 0, and
after the commit titled "usb-ehci: frindex always is a 14 bits counter"
we rely on frindex always being a multiple of 8. I've not seen this in
practice, but theoretically a guest can write a value >= 0x4000 or a value
which is not a multiple of 8 value to frindex, this patch ensures that
things will still work when that happens.
To answer the question in the comment removed by this patch: I think
this was needed because several places in the ehci emulation did not
check the T bit of link entries correctly and thus might have followed
invalid references. See commit 2a5ff735dc1074171a0cbb1dc228d6d6e907f571
It's identical to the hda-duplex codec, except that it advertises the
input as microphone instead of line-in and the output as speaker instead
of line-out. Some guest apps (microsoft netmeeting being one) are picky
when it comes to selecting the recording source and don't accept
line-in, so give them what they expect.
Amit Shah [Tue, 24 Apr 2012 14:03:59 +0000 (19:33 +0530)]
virtio-serial-bus: Unset guest_connected at reset and driver reset
When a guest driver resets the virtio status to not ready, or when qemu
is reset, reset all ports' guest_connected bit and let port users know
of this event if they have the guest_close() callback registered.
Alon Levy [Mon, 23 Apr 2012 15:23:10 +0000 (18:23 +0300)]
virtio-serial-bus: fix guest_connected init before driver init
guest_connected should be false before guest driver initialization, and
true after, both for multiport aware and non multiport aware drivers.
Don't set it before the guest_features are available; instead use
set_status which is called by io to VIRTIO_PCI_STATUS with
VIRTIO_CONFIG_S_DRIVER_OK by even older non multiport drivers.
[Amit: Add comment, tweak summary, only set guest_connected and not
reset it as a side-effect.]
Andreas Färber [Tue, 17 Apr 2012 17:22:58 +0000 (19:22 +0200)]
target-i386: Prepare "vendor" property for X86CPU
Using it now would incur converting the three x86_def_t vendor words
into a string for object_property_set_str(), then back to three words
in the "vendor" setter.
The built-in CPU definitions use numeric preprocessor defines to
initialize the three words in a charset-safe way, so do not change the
fields to char[12] just to use the setter.
Andreas Färber [Tue, 17 Apr 2012 14:15:15 +0000 (16:15 +0200)]
target-i386: Add range check for -cpu ,family=x
A family field value of 0xf and extended family field value of 0xff is
the maximum representable unsigned family number.
All other CPUID property values are bounds-checked, so add a check here
for symmetry before we adopt it in a property setter.
Jason Wang [Thu, 22 Mar 2012 10:02:24 +0000 (18:02 +0800)]
e1000: link auto-negotiation emulation
Indeed, there's nothing else except for the time spent on the
negotiation needs to be emulated. This is needed for resuming windows
guest from hibernation, as without a proper delay, qemu would send the
packet too early ( guest even does not have a proper intr handler),
which could lead windows guest hang.
This patch first introduces an array of function pointers to make it
possible to emulate per-register write behavior. Then traps the
PHY_CTRL register write and when guest want to restart the link auto
negotiation, we would down the link and mark the auto negotiation in
progress in PHY_STATUS register. After time, a timer with 500 ms (
which is the minimum timeout of auto-negotation specified in 802.3
spec). The link would be up when timer expired.
Test with resuming windows guest plus flood ping and linux ethtool
linkstatus test.
Jason Wang [Thu, 22 Mar 2012 10:02:07 +0000 (18:02 +0800)]
e1000: introduce helpers to manipulate link status
This patch introduces helpers to change link status bit for phy/mac
register. This would help to reduce code duplication and would be used
by following patches.
Stefan Weil [Tue, 10 Apr 2012 18:48:54 +0000 (20:48 +0200)]
eepro100: Fix multicast regression
Commit 7fc8d918b9674c3e9233d6d25da2457345d414a0 removed code from
eepro100.c and replaced it by different code: the code in net.c
returns bits 31...26, but eepro100 needs bits 7...2.
if (vq->last_avail_index != vring_avail_idx(vq)) {
read descriptor head at vq->last_avail_index;
}
In theory, processor can reorder descriptor head
read to happen speculatively before the index read.
this would trigger the following race:
host descriptor head read <- reads invalid head from ring
guest writes valid descriptor head
guest writes avail index
host avail index read <- observes valid index
as a result host will use an invalid head value.
This was not observed in the field by me but after
the experience with the previous two races
I think it is prudent to address this theoretical race condition.
This fixes an issue dual to the one fixed by
patch 'virtio: add missing mb() on notification'
and applies on top.
In this case, to enable vq kick to exit to host,
qemu writes out used flag then reads the
avail index. if these are reordered we get a race:
host avail index read: ring is empty
guest avail index write
guest flag read: exit disabled
host used flag write: enable exit
which results in a lost exit: host will never be notified about the
avail index update. Again, happens in the field but only seems to
trigger on some specific hardware.
Insert an smp_mb barrier operation to ensure the correct ordering.
During normal operation, virtio first writes a used index
and then checks whether it should interrupt the guest
by reading guest avail index/flag values.
Guest does the reverse: writes the index/flag,
then checks the used ring.
The ordering is important: if host avail flag read bypasses the used
index write, we could in effect get this timing:
host avail flag read
guest enable interrupts: avail flag write
guest check used ring: ring is empty
host used index write
which results in a lost interrupt: guest will never be notified
about the used ring update.
This actually can happen when using kvm with an io thread,
such that the guest vcpu and qemu run on different host cpus,
and this has actually been observed in the field
(but only seems to trigger on very specific processor types)
with userspace virtio: vhost has the necessary smp_mb()
in place to prevent the regordering, so the same workload stalls
forever waiting for an interrupt with vhost=off but works
fine with vhost=on.
Insert an smp_mb barrier operation in userspace virtio to
ensure the correct ordering.
Applying this patch fixed the race condition we have observed.
Tested on x86_64. I checked the code generated by the new macro
for i386 and ppc but didn't run virtio.
Note: mb could in theory be implemented by __sync_synchronize, but this
would make us hit old GCC bugs. Besides old GCC
not implementing __sync_synchronize at all, there were bugs
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36793
in this functionality as recently as in 4.3.
As we need asm for rmb,wmb anyway, it's just as well to
use it for mb.
Peter Chubb [Fri, 20 Apr 2012 05:32:30 +0000 (15:32 +1000)]
Limit ptimer rate to something achievable
If a guest sets very short timeouts, and asks for a timer to be reloaded on
timeout, QEMU can go to 100%CPU utilisation and become unresponsive,
as it is spending all its time generating timeout interrupts. On real
hardware this doesn't matter, as the interrupts are just coalesced,
and the effect is to have the interrupt asserted all the time.
This patch is a band-aid, that prevents timeouts less than 10
microseconds from being set. 10 microseconds is a limit that was
determined empirically on a variety of machines as the shortest that
allowed QEMU to pick up a control-a c sequence to get at the monitor.
Stefan Weil [Thu, 19 Apr 2012 20:27:14 +0000 (22:27 +0200)]
qemu-char: Fix crash when switching consoles
qemu-system-arm (and other system emulations) crashes with SDL when
the user switches consoles (Alt-Ctrl-F4).
We already check for NULL pointers in qemu_chr_fe_ioctl,
qemu_chr_be_can_write and other functions, so do this also
for s->chr_read in qemu_chr_be_write. This fixes the crash.
configure: change meaning of --datadir to Autoconf convention
Autoconf uses --datadir for the /usr/share directory, not the
program-specific subdirectory inside /usr/share. This changes configure
to match autoconf behavior.
Note that this will break compatibility with existing build scripts.
As now that block handles only the prefix variable, the code can be much
simpler. This also removes the CONFIG_QEMU_PREFIX define as it is not
used by any C code.
This one will be kept because it is set directly by ./configure:
- CONFIG_QEMU_HELPERDIR
This patch keeps the 'prefix=*' (CONFIG_QEMU_PREFIX) pattern because
other variables may use $prefix on their config-host.mak definitions.
The remaining code will be simplified on a further patch.
$confdir is a confusing name, as it's not clear if it's "the system
config dir" or "the Qemus-specific directory inside the config dir".
$qemu_confdir makes it more clear.
The current C code that uses CONFIG_QEMU_CONFDIR should be safe, as now
create_config generates the same #define name (CONFIG_QEMU_CONFDIR) for
both "confdir" and "qemu_confdir" variables.
Autoconf concept of "datadir" is supposed to be "$prefix/share", not
"$prefix/share/PACKAGE", so using datadir for the Qemu-specific
directory is confusing.
The current C code that uses CONFIG_QEMU_DATADIR should be safe, as now
create_config generates the same #define name (CONFIG_QEMU_DATADIR) for
both "datadir" and "qemu_datadir" variables.
create_config: separate section for qemu_*dir variables (v2)
The generic *dir section will eventually go away and be replaced with
qemu_* section. By now, both sections will be kept, while the variables
get renamed on config-host.mak.
With this patch, a XXXdir variable will become a CONFIG_QEMU_XXXDIR
define, and a qemu_XXXdir variable will become CONFIG_QEMU_XXXDIR as
well (instead of becoming a CONFIG_QEMU_QEMU_XXXDIR define).
Changes v1 -> v2:
- Rebase on top of newer qemu.git changes, that changed
"tr '[:lower:]' '[:upper:]'" to "LC_ALL=C tr '[a-z]' '[A-Z]'".
The autoconf "docdir" variable is used for the program-specific
subdirectory, but anyway it's better to make it clear that the variable
is used for the qemu-specific subdirectory path.
The variable is used for the qemu-specific directory and has a different
meaning of the autoconf "datadir" variable (that's used for the
$prefix/share directory, not for $prefix/share/PACKAGE).
This doesn't change behavior or interfaces, it's just an internal
variable rename.
Andreas Färber [Thu, 12 Apr 2012 00:17:53 +0000 (02:17 +0200)]
target-microblaze: QOM'ify CPU
Embed CPUMBState as first member of QOM MicroBlazeCPU.
Let CPUClass::reset() call cpu_state_reset() for now.
Signed-off-by: Andreas Färber <[email protected]> Tested-by: Peter A. G. Crosthwaite <[email protected]>
[AF: Updated cpu.c to include cpu-qom.h indirectly via cpu.h]
projects / qemu.git / log