aliguori [Thu, 15 Jan 2009 21:39:41 +0000 (21:39 +0000)]
Fix Windows build (Hervé Poussineau)
r6303 broke Windows build, where "noreturn" is a keyword used with __declspec.
Attached patch fixes Windows build, by moving windows.h header inclusion before Qemu noreturn define.
aliguori [Thu, 15 Jan 2009 20:47:45 +0000 (20:47 +0000)]
Handle SDL grabs failing (Mark McLoughlin)
If a X window is not viewable, XGrabPointer() fails and
returns GrabNotViewable. SDL's X backend currently handles
this by retrying the grab until the window becomes viewable
again.
This means e.g. if you Ctrl-Alt-RightArrow to switch
workspaces, QEMU tries to grab, SDL blocks because the
window isn't viewable and your guest stops executing until
you switch back to that workspace again.
See this Fedora bug for the gory details:
https://bugzilla.redhat.com/480065
Some SDL backends will return SDL_GRAB_OFF from
SDL_WM_GrabInput(), so the fix is to make the X backend do
this if the grab fails.
The only side-effect in QEMU is that if SDL_WM_GrabInput()
fails we still change the window title to indicate that it's
grabbed, when in fact it's not. This patch fixes that minor
issue.
aliguori [Thu, 15 Jan 2009 20:37:28 +0000 (20:37 +0000)]
add a -vga none cli option (Stefano Stabellini)
currently there is no way to fully disable any graphic card device for
the PC architecture.
You can have no graphical output, thanks to -nographic, but you would
have the VGA device connected to your PCI bus anyway.
There is already a convenient -vga option to choose between std, cirrus
and vmware; this patch add the new option "none" to select no graphic
card at all.
aliguori [Thu, 15 Jan 2009 20:11:34 +0000 (20:11 +0000)]
Add -rtc-td-hack option to fix time drift with RTC on Windows (Gleb Natapov)
After my last patch to fix interrupt coalescing was rejected
on the basis that it is too intrusive we decided to make the
fix much more localized and only fix the problem for RTC time
source. Unfortunately it is impossible to fix the problem entirely
inside RTC code like Andrzej proposed since Windows reads RTC
register C more then once on each time interrupt so it is impossible
to count reliably how many interrupt windows actually handled.
Proposed solution is localized to I386 target and is disabled by
default. To enable it "-rtc-td-hack" flag should be used.
This patch adds the typical qemu console command line switch to the virtio
console. using -virtioconsole ARG it can now be specified what output a guest
hvc should be redirected to.
aliguori [Thu, 15 Jan 2009 20:05:25 +0000 (20:05 +0000)]
add virtio-console support (Christian Ehrhardt)
This patch adds the virtio console to qemu. This console can be found after the
serial and parallel outputs as another virtual console. In the -nographic case
it is redirected to the null output by default.
aurel32 [Wed, 14 Jan 2009 21:09:07 +0000 (21:09 +0000)]
Fix day of week in mc146818
According to mc146818 specification, Day of Week register (#6) is between 1
and 7, 1 representing Sunday. According C specification, tm_wday field in
struct tm is between 0 and 6, 0 representing Sunday.
Bit 2 of register B (#11) is named DM (data mode) and specifies if RTC
stores values in BCD or in binary form.
blueswir1 [Wed, 14 Jan 2009 19:00:36 +0000 (19:00 +0000)]
Add noreturn function attribute
Introduce noreturn attribute and attach it to cpu_loop_exit as well as
interrupt/exception helpers for i386. This avoids a bunch of gcc4
warnings.
[ Note that this patch comes with a workaround to include qemu-common.h
even in cases where is currently causes conflicts with dyngen-exec.h.
I've been told that these conflicts will get resolved in the future
(/me will try to have a look as well - as time permits). ]
malc [Wed, 14 Jan 2009 18:39:49 +0000 (18:39 +0000)]
Remove all traces of __powerpc__
According to $GCC/gcc/config/rs6000/rs6000-c.c _ARCH_PPC is the
ubiquitous define which should be used to test whether gcc targets
PowerPC, on 64bit platforms _ARCH_PPC64 will be also defined.
aurel32 [Wed, 14 Jan 2009 14:47:56 +0000 (14:47 +0000)]
escc: allow one IRQ per serial channel
The Z85C30 on the PowerMAC machines have one interrupt per serial
channel, while the Sparc machines have only one for both. Allow the
emulated device to use one IRQ per channel.
aliguori [Tue, 13 Jan 2009 19:39:36 +0000 (19:39 +0000)]
Check NIC model in some NIC init functions (Mark McLoughlin)
Some NIC init functions are only called when that model is
the only valid model. In that case, it makes sense to use
qemu_check_nic_model() from the NIC init function itself.
aliguori [Tue, 13 Jan 2009 19:15:55 +0000 (19:15 +0000)]
Fix tap downscript argument (Mark McLoughlin)
Kill off the hack that parses info_str for the tap interface
name to pass as the argument to the downscript and, instead,
just explicitly keep a copy of the string for later.
aliguori [Tue, 13 Jan 2009 15:20:14 +0000 (15:20 +0000)]
RTL8139: Latch C+ mode state instead of inferring it from C+ Command register (Avi Kivity)
It was observed that Windows 2003 x64 hangs when shutting down if an
RTL8139 NIC and a USB device tablet are both present. What seems to be
happening is:
- the guest shuts down the transmitter and receiver
- time passes
- the guest requests a tally counter dump
As it happens, the tally counter command register overlaps the transmit
status register in C mode. Qemu determines whether the chip is in C or C+
mode by looking at the C+ transmit enable bit; as this is now unset, the
dump tally counter command is interpreted as a C mode transmit command. The
guest doesn't think so, however, and continues to poll for completion of the
tally counter dump command. This never occurs, so the guest hangs.
Fix by redefining C+ mode as "a write to the C+ command register has occurred
since the last reset". The data sheet is silent on the matter.
aliguori [Tue, 13 Jan 2009 15:13:53 +0000 (15:13 +0000)]
Fix race in POSIX AIO emulation (Jan Kiszka)
When we cancel an AIO request that is already being processed by
aio_thread, qemu_paio_cancel should return QEMU_PAIO_NOTCANCELED as long
as aio_thread isn't done with this request. But as the latter currently
updates aiocb->ret after every block of the request, we may report
QEMU_PAIO_ALLDONE too early.
Futhermore, in case some zero-length request should have been queued,
aiocb->ret is never set to != -EINPROGRESS and callers like
raw_aio_cancel could get stuck in an endless loop.
Fix those issues by updating aiocb->ret _after_ the request has been
fully processed. This also simplifies the locking.
aliguori [Fri, 9 Jan 2009 20:05:10 +0000 (20:05 +0000)]
fix configuring kvm probe when using --kerneldir (Christian Ehrhardt)
There is already a variable kvm_cflags which gets the path of the kernel
includes when using --kerneldir. But eventually with newer kernels we all will
need arch/$arch/include too (my case was a incldue of asm/kvm.h which was not
found anymore). Headers in a full kernel source are not flattened to
one arch like they are if e.g. installed kernel headers are used.
To fix that, the includes added to cflags depending on --kerneldir should also
contian the arch includes. The patch adds a special check for x86 because its
source layout recently changed, all others directly use arch/$cpu/include if
existent.
aliguori [Thu, 8 Jan 2009 21:03:55 +0000 (21:03 +0000)]
Enable ac97 by default
ac97 has drivers for Vista 64-bit whereas sb16 and es1370 do not appear to. I
asked malc why it was disabled and he said it was because it was GPL. He did
not object to enabling it now that more QEMU code is GPL'd.
aliguori [Thu, 8 Jan 2009 21:03:07 +0000 (21:03 +0000)]
IDE: Return zero when reading error register with slave selected (Justin Chevrier)
During hardware detection Openserver issues commands to slaves on both primary
and secondary ports. We already return a zero for the status register in this
situation but after reading the status register the Openserver installer
proceeds to check the value of the error register. Currently we return the
existing value in the register. This confuses the installer and it tries to
access the slave units later on. When the command that gets issued later gets
ignored the installer freezes. The patch below returns zero when reading the
error register if a slave unit is selected. Openserver can now successfully be
installed using the emulated IDE hard drive.
Return zero when reading error register with slave selected
aliguori [Thu, 8 Jan 2009 19:45:50 +0000 (19:45 +0000)]
Implement e1000 link status (Mark McLoughlin)
On link up or down we set the E1000_STATUS_LU ("link up") bit
in the status register and set the E1000_ICR_LSC ("link
status changed") bit in the interrupt cause register before
interrupting the guest.
aliguori [Thu, 8 Jan 2009 19:01:37 +0000 (19:01 +0000)]
Add model field to nic info (Mark McLoughlin)
The model type used to be printed as part of the nic info. It was removed when
the name type was added. This adds back a model field for those that were
using it previously.
projects / qemu.git / log