Peter Maydell [Thu, 19 May 2011 13:46:16 +0000 (14:46 +0100)]
target-arm: Signal InvalidOp for Neon GE and GT compares of QNaN
If the input to a Neon float comparison is a quiet NaN, the ARM ARM
specifies that we should raise InvalidOp if the comparison is GE or GT
but not for EQ. (Signaling NaNs raise InvalidOp regardless). This means
only EQ should use the _quiet version of the comparison function.
We implement this by cleaning up the comparison helpers to call the
appopriate versions of the softfloat simple comparison functions
(float32_le and friends) rather than the generic float32_compare functions.
This makes them simple enough that they are clearer opencoded rather
than macroised.
Peter Maydell [Thu, 19 May 2011 13:46:19 +0000 (14:46 +0100)]
target-arm: Use correct float status for Neon int-float conversions
The Neon versions of int-float conversions must use the "standard FPSCR"
rather than the default FPSCR. Implement this by having the helper
functions take a pointer to the appropriate float_status value rather
than simply taking a pointer to the entire CPUState, and making
translate.c pass a pointer to vfp.fp_status or vfp.standard_fp_status
appropriately for whether the instruction being translated is Neon
or VFP.
Peter Maydell [Thu, 19 May 2011 13:46:18 +0000 (14:46 +0100)]
target-arm: Signal Underflow when denormal flushed to zero on output
On ARM the architecture mandates that when an output denormal is flushed to
zero we must set the FPSCR UFC (underflow) bit, so map softfloat's
float_flag_output_denormal accordingly.
Peter Maydell [Thu, 19 May 2011 13:46:17 +0000 (14:46 +0100)]
softfloat: Add new flag for when denormal result is flushed to zero
Add a new float_flag_output_denormal which is set when the result
of a floating point operation would be denormal but is flushed to
zero because we are in flush_to_zero mode. This is necessary because
some architectures signal this condition as an underflow and others
signal it as an inexact result.
Peter Maydell [Thu, 19 May 2011 13:46:15 +0000 (14:46 +0100)]
target-arm: Signal InputDenormal for VRECPE, VRSQRTE, VRECPS, VRSQRTS
The helpers for VRECPE.F32, VSQRTE.F32, VRECPS and VRSQRTS handle denormals
as special cases, so we must set the InputDenormal exception flag ourselves.
Peter Maydell [Thu, 19 May 2011 13:46:14 +0000 (14:46 +0100)]
target-arm: Don't set FP exceptions in recip, recip_sqrt estimate fns
The functions which do the core estimation algorithms for the VRSQRTE
and VRECPE instructions should not set floating point exception flags,
so use a local fp status for doing these calculations.
Aurelien Jarno [Tue, 17 May 2011 16:25:45 +0000 (18:25 +0200)]
tcg: don't keep dead outputs in registers
If an op with dead outputs is not removed, because it has side effects
or has multiple output and only one dead, mark the registers as dead
instead of saving them. This avoid a few register spills on TCG targets
with low register count, especially with div2 and mul2 ops, or when a
qemu_ld* result is not used (prefetch emulation for example).
Aurelien Jarno [Mon, 23 May 2011 20:35:37 +0000 (22:35 +0200)]
Merge branch 'ppc-next' of git://repo.or.cz/qemu/agraf
* 'ppc-next' of git://repo.or.cz/qemu/agraf:
Fix a bug in mtsr/mtsrin emulation on ppc64
pSeries: Clean up write-only variables
w32: Fix compilation and replace non-portable usage of ulong
Aurelien Jarno [Mon, 23 May 2011 20:33:39 +0000 (22:33 +0200)]
Merge branch 's390-next' of git://repo.or.cz/qemu/agraf
* 's390-next' of git://repo.or.cz/qemu/agraf:
s390x: complain when allocating ram fails
s390x: fix memory detection for guests > 64GB
s390x: change mapping base to allow guests > 2GB
s390x: Fix debugging for unknown sigp order codes
s390x: build s390x by default
s390x: remove compatibility cc field
s390x: Adjust GDB stub
s390x: translate engine for s390x CPU
s390x: Adjust internal kvm code
s390x: Implement opcode helpers
s390x: helper functions for system emulation
s390x: Shift variables in CPUState for memset(0)
s390x: keep hint on virtio managing size
s390x: make kvm exported functions conditional on kvm
s390x: s390x-linux-user support
tcg: extend max tcg opcodes when using 64-on-32bit
s390x: fix smp support for kvm
Peter Maydell [Fri, 20 May 2011 09:11:53 +0000 (10:11 +0100)]
hw/sd.c: Don't complain about SDIO commands CMD52/CMD53
The SDIO specification introduces new commands 52 and 53.
Handle as illegal command but do not complain on stderr,
as SDIO-aware OSes (including Linux) may legitimately use
these in their probing for presence of an SDIO card.
Blue Swirl [Sat, 14 May 2011 11:55:30 +0000 (11:55 +0000)]
Delete unused tb_invalidate_page_range
tb_invalidate_page_range() was intended to be used to invalidate an
area of a TB which the guest explicitly flushes from i-cache. However,
QEMU detects writes to code areas where TBs have been generated, so
his has never been useful.
David Gibson [Fri, 20 May 2011 03:34:59 +0000 (13:34 +1000)]
Fix a bug in mtsr/mtsrin emulation on ppc64
Early ppc64 CPUs include a hack to partially simulate the ppc32 segment
registers, by translating writes to them into writes to the SLB. This is
not used by any current Linux kernel, but it is used by the openbios used
in the qemu mac99 model.
Commit 81762d6dd0d430d87024f2c83e9c4dcc4329fb7d, cleaning up the SLB
handling introduced a bug in this code, breaking the openbios currently in
qemu. Specifically, there was an off by one error bitshuffling the
register format used by mtsr into the format needed for the SLB load,
causing the flag bits to end up in the wrong place. This caused the
storage keys to be wrong under openbios, meaning that the translation code
incorrectly thought a legitimate access was a permission violation.
This patch fixes the bug, at the same time it fixes some build bug in the
MMU debugging code (only exposed when DEBUG_MMU is enabled).
David Gibson [Tue, 17 May 2011 06:47:04 +0000 (16:47 +1000)]
pSeries: Clean up write-only variables
A few pieces of the pSeries emulation code have variables which are set
but never used, which causes warnings on gcc 4.6. This patch removes
these instances.
Stefan Weil [Sat, 14 May 2011 23:38:04 +0000 (23:38 +0000)]
w32: Fix compilation and replace non-portable usage of ulong
ulong is undefined for w32 (and maybe other) compilations.
Replace it by uintptr_t (which also fixes compilation for w64
and is a better choice for pointer to integer conversions).
Alexander Graf [Fri, 20 May 2011 15:33:28 +0000 (17:33 +0200)]
s390x: complain when allocating ram fails
While trying out the > 64GB guest RAM patch, I hit some virtual address
limitations of my host system, which resulted in mmap failing. Unfortunately,
qemu didn't tell me about this failure, but just used the NULL pointer
happily, resulting in either segmentation faults or other fun errors.
To spare other users from tracing this down, let's print a nice message
instead so the user can figure out what's wrong from there.
the s390 memory detection has a 16bit field that specifies the amount of
increments. This patch adopts the memory size to always fit into that
scheme. This also fixes virtio detection for these guests, since the
descriptor page is located after the main memory.
which limits the amount of available memory to <2GB.
This patch moves the guest pages to 32GB to not collide with the binary
and to leave some space for the program break of qemu.
On unknown sigp order codes we print a debug message. This patch
fixes the output, since we want to see the order_code and not
the register numbers.
Patch applies on agraf tree.
Alexander Graf [Wed, 23 Mar 2011 09:58:07 +0000 (10:58 +0100)]
s390x: Adjust GDB stub
We have successfully lazilized cc computation, so we need to manually
trigger its calculation when gdb wants to fetch it. We also changed the
variable name, so writing it writes into a different field now.
Alexander Graf [Mon, 2 May 2011 08:11:40 +0000 (10:11 +0200)]
s390x: Shift variables in CPUState for memset(0)
The default reset handler does a memset(0) until right in between CPU_COMMON.
I incorrectly changed that behavior on the s390x port, so let's move the fields
in CPUState around to reflect the correct split up to which point memset(0)
zeros out everything.
Alexander Graf [Wed, 13 Apr 2011 08:55:11 +0000 (10:55 +0200)]
s390x: keep hint on virtio managing size
The s390x virtio bus keeps management information on virtio after the top
of the guest's RAM. We need to be able to tell the guest the size of its
RAM (without virtio stuff), but also be able to trap when the guest accesses
RAM outside of its scope (including virtio stuff).
So we need a variable telling us the size of the virtio stuff, so we can
calculate the highest available RAM address from that.
While at it, also increase the maximum number of virtio pages, so we play
along well with more recent kernels that spawn a ridiculous number of virtio
console adapters.
Currently smp support for kvm does not work. Qemu does a kvm run even on
secondary CPUs which dont have a sane state (initial psw == 0)
triggering some program faults. Architecturally these cpus are in the stopped
state, so we should not do the kvm run ioctl. (these CPUs will be started
by a SIGP restart later during the boot process)
We need to tell the loop that this cpu should not run. Jan Kiszka pointed
out that kvm_arch_process_async_events is the right place to do.
Alexander Graf [Thu, 19 May 2011 09:57:09 +0000 (11:57 +0200)]
ahci: Fix non-NCQ accesses for LBA > 16bits
AHCI provides two ways of reading/writing data:
1) NCQ
2) ATA commands with the LBA in the command FIS
In the second code path, we didn't handle any LBAs that were bigger than
16 bits, so whenever a guest that used high LBA numbers wanted to access
data, the LBA got truncated down to 16 bits, giving the guest garbage.
This patch adds support for LBAs higher than 16 bits. I've tested that it
works just fine with SeaBIOS and Linux guests. This patch also unbreaks
the often reported grub errors people have seen with AHCI.
Ignore pci unplug requests for unpluggable devices (CVE-2011-1751)
This patch makes qemu ignore unplug requests from the guest for pci
devices which are tagged as non-hotpluggable. Trouble spot is the
piix4 chipset with the ISA bridge. Requests to unplug that one will
make it go away together with all ISA bus devices, which are not
prepared to be unplugged and thus don't cleanup, leaving active
qemu timers behind in free'ed memory.
DriveInfo is closely tied to -drive, and like -drive, it mixes
information about host and guest part of the block device. Unlike
DriveInfo, BlockDriverState should be about the host part only.
One of the remaining guest bits there is the "type hint". -drive
option media sets it, and qdevs "ide-drive", "scsi-disk" and non-qdev
IF_XEN devices check it to pick HD vs. CD.
Communicate -drive option media via new DriveInfo member media_cd
instead.
block QMP: Deprecate query-block's "type", drop info block's "type="
query-block's specification documents response member "type" with
values "hd", "cdrom", "floppy", "unknown".
Its value is unreliable: a block device used as floppy has type
"floppy" if created with if=floppy, but type "hd" if created with
if=none.
That's because with if=none, the type is at best a declaration of
intent: the drive can be connected to any guest device. Its type is
really the guest device's business. Reporting it here is wrong.
No known user of QMP uses "type". It's unlikely that any unknown
users exist, because its value is useless unless you know how the
block device was created. But then you also know the true value.
Fixing the broken value risks breaking (hypothetical!) clients that
somehow rely on the current behavior. Not fixing the value risks
breaking (hypothetical!) clients that rely on the value to be
accurate. Can't entirely avoid hypothetical lossage. Change the
value to be always "unknown".
This makes "info block" always report "type=unknown". Pointless.
Change it to not report the type.
scsi: Split qdev "scsi-disk" into "scsi-hd" and "scsi-cd"
A "scsi-disk" is either a hard disk or a CD-ROM, depending on the
associated BlockDriverState's type hint. Unclean; disk vs. CD belongs
to the guest part, not the host part.
Have separate qdevs "scsi-hd" and "scsi-cd" to model disk vs. CD in
the guest part.
Keep scsi-disk for backward compatibility.
Don't copy scsi-disk property removable to scsi-cd. It's not used and
always zero(!) there.
ide: Split qdev "ide-drive" into "ide-hd" and "ide-cd"
An "ide-drive" is either a hard disk or a CD-ROM, depending on the
associated BlockDriverState's type hint. Unclean; disk vs. CD belongs
to the guest part, not the host part.
Have separate qdevs "ide-hd" and "ide-cd" to model disk vs. CD in
the guest part.
Keep ide-drive for backward compatibility.
"ide-disk" would perhaps be a nicer name than "ide-hd", but there's
already "scsi-disk", which is like "ide-drive", and will be likewise
split in the next commit. {ide,scsi}-{hd,cd} is the best consistent
set of names I could find within the backward compatibility
straightjacket.
Stefan Hajnoczi [Mon, 16 May 2011 12:56:53 +0000 (13:56 +0100)]
qed: support for growing images
The .bdrv_truncate() operation resizes images and growing is easy to
implement in QED. Simply check that the new size is valid and then
update the image_size header field to reflect the new size.
I run qemu-img under profiler and realized, that most of CPU time is
consumed by is_not_zero() function. I had made a couple of optimizations
on it and got the following output for `time qemu-img convert -O qcow2
volume.qcow2 snapshot.qcow2`:
Original qemu-img:
real 0m56.159s
user 0m34.670s
sys 0m12.079s
Patched qemu-img:
real 0m34.805s
user 0m18.445s
sys 0m12.552s
Stefan Hajnoczi [Mon, 9 May 2011 15:45:40 +0000 (16:45 +0100)]
qed: Periodically flush and clear need check bit
One strategy to limit the startup delay of consistency check when
opening image files is to ensure that the file is marked dirty for as
little time as possible.
QED currently marks the image dirty when the first allocating write
request is issued and clears the dirty bit again when the image is
cleanly closed. In practice that means the image is marked dirty for
most of a guest's lifetime and prone to being in a dirty state upon
crash or power failure.
It is safe to clear the dirty bit after all allocating write requests
have completed and a flush has been performed. This patch adds a timer
after the last allocating write request completes. When the timer fires
it will flush and then clear the dirty bit. The timer is set to 5
seconds and is cancelled upon arrival of a new allocating write request.
Jan Kiszka [Mon, 9 May 2011 15:48:19 +0000 (17:48 +0200)]
ahci: Fix crashes on duplicate BH registration
If ahci_dma_set_inactive is called a while there is still a pending BH
from a previous run, we will crash on the second run of
ahci_check_cmd_bh as it overwrites AHCIDevice::check_bh. Avoid this
broken and redundant duplicate registration.
Kevin Wolf [Mon, 9 May 2011 09:42:03 +0000 (11:42 +0200)]
ide: Turn debug messages into assertions
These printfs aren't really debug messages, but clearly indicate a bug if they
ever become effective. Noone uses DEBUG_IDE, let's re-enable the check
unconditionally and make it an assertion instead of printfs in the device
emulation.
Kevin Wolf [Mon, 2 May 2011 15:32:54 +0000 (17:32 +0200)]
posix-aio-compat: Fix idle_threads counter
A thread should only be counted as idle when it really is waiting for new
requests. Without this patch, sometimes too few threads are started as busy
threads are counted as idle.
Not sure if it makes a difference in practice outside some artificial
qemu-io/qemu-img tests, but I think the change makes sense in any case.
Stefan Weil [Sat, 7 May 2011 20:49:33 +0000 (22:49 +0200)]
target-lm32: Remove unused local variables
cppcheck report:
target-lm32/translate.c:587: style:
Variable 't0' is assigned a value that is never used
target-lm32/translate.c:588: style:
Variable 'l1' is assigned a value that is never used
Remove both variables. Please check whether that is the correct solution.
Peter Maydell [Thu, 5 May 2011 18:35:35 +0000 (19:35 +0100)]
target-arm: Fix VMLA, VMLS, VNMLS, VNMLA handling of NaNs
Correct handling of NaNs for VFP VMLA, VMLS, VNMLS and VNMLA requires that
we implement the set of negations and additions specified by the ARM ARM;
plausible looking simplifications like turning (-A + B) into (B - A) or
computing (A + B) rather than (B + A) result in selecting the wrong NaN or
returning a NaN with the wrong sign bit.
doc: Add explanation that -alt-grab and -ctrl-grab affect special keys
Phillip Merensky reported that the special keys (e.g. Ctrl-Alt-f for full
screen) did not work correctly if -alt-grab is used.
BUG: 696530
Review of ui/sdl.c:sdl_refresh indicates that this is the intended behaviour,
so we should update the documentation to match the actual behaviour, as
suggested by Phillip in the bug report.
Aurelien Jarno [Sat, 14 May 2011 14:54:59 +0000 (16:54 +0200)]
Merge branch 'ppc-next' of git://repo.or.cz/qemu/agraf
* 'ppc-next' of git://repo.or.cz/qemu/agraf:
PPC: Qdev'ify e500 pci
PPC MPC7544DS: Use new TLB helper function
PPC: Implement e500 (FSL) MMU
PPC: Add another 64 bits to instruction feature mask
PPC: Add GS MSR definition
PPC: Make MPC8544DS emulation work w/o KVM
PPC: Make MPC8544DS obey -cpu switch
Fix off-by-one error in sizing pSeries hcall table
ppc64: Fix out-of-tree builds
kvm: ppc: warn user on PAGE_SIZE mismatch
kvm: ppc: detect old headers
monitor: add PPC BookE SPRs
kvm: ppc: fixes for KVM_SET_SREGS on init
ppc64: Don't try to build sPAPR RTAS on Darwin
Place pseries vty devices at addresses more similar to existing machines
Make pSeries 'model' property more closely resemble real hardware
pseries: Increase maximum CPUs to 256
Alexander Graf [Fri, 6 May 2011 08:37:56 +0000 (10:37 +0200)]
PPC MPC7544DS: Use new TLB helper function
Now that we have some nice helpers that can find us a TLB entry, let's
use that on the machine initialization code, so we don't need to know
about the internals of the TLB array.
Alexander Graf [Sat, 30 Apr 2011 21:34:58 +0000 (23:34 +0200)]
PPC: Implement e500 (FSL) MMU
Most of the code to support e500 style MMUs is already in place, but
we're missing on some of the special TLB0-TLB1 handling code and slightly
different TLB modification.
Alexander Graf [Sat, 30 Apr 2011 22:00:58 +0000 (00:00 +0200)]
PPC: Add another 64 bits to instruction feature mask
To enable quick runtime detection of instruction groups to the currently
selected CPU emulation, we have a feature mask of what exactly the respective
instruction supports.
This feature mask is 64 bits long and we just successfully exceeded those 64
bits. To add more features, we need to think of something.
The easiest solution that came to my mind was to simply add another 64 bits
that we can also match on. Since the comparison is only done on start of the
qemu process to generate an internal opcode calling table, we should be fine
on any performance penalties here.
Alexander Graf [Sat, 30 Apr 2011 21:05:03 +0000 (23:05 +0200)]
PPC: Make MPC8544DS obey -cpu switch
The MPC8544DS board emulation code ignored the user defined -cpu switch.
This patch enables it to only provide a sane default, not force an e500v2
CPU inside.
David Gibson [Tue, 10 May 2011 06:06:21 +0000 (16:06 +1000)]
Fix off-by-one error in sizing pSeries hcall table
The pSeries machine uses two tables to look up guest hcalls for emulation.
One of these is exactly one entry too small to hold all the hcalls it needs
to, leading to memory corruption.
This patch fixes the bug, and while we're at it, make both tables 'static'
since they're never used from other modules.
Alexander Graf [Sat, 16 Apr 2011 08:15:11 +0000 (10:15 +0200)]
kvm: ppc: warn user on PAGE_SIZE mismatch
On PPC, the default PAGE_SIZE is 64kb. Unfortunately, the hardware
alignments don't match here: There are RAM and MMIO regions within
a single page when it's 64kb in size.
So the only way out for now is to tell the user that he should use 4k
PAGE_SIZE.
This patch gives the user a hint on that, telling him that failing to
register a prefix slot is most likely to be caused by mismatching PAGE_SIZE.
This way it's also more future-proof, as bigger PAGE_SIZE can easily be
supported by other machines then, as long as they stick to 64kb granularities.
Alexander Graf [Sat, 16 Apr 2011 00:00:36 +0000 (02:00 +0200)]
kvm: ppc: detect old headers
When compiling Qemu with older kernel headers, the PVR setting
mechanism isn't available yet. Unfortunately, back then I didn't add
a capability we could check against, so all we can do is add a configure
test to see if we support PVR setting. For BookE, we don't care yet.
This fixes compilation errors with KVM enabled on older kernel headers
(like 2.6.32).
Scott Wood [Mon, 11 Apr 2011 23:34:34 +0000 (18:34 -0500)]
kvm: ppc: fixes for KVM_SET_SREGS on init
Classic/server ppc has had SREGS for a while now (though I think not
always?), but it's still missing for booke. Check the capability before
calling KVM_SET_SREGS.
Also, don't write random stack state into the non-PVR sregs fields --
have kvm fill it in first.
Eventually booke will have sregs and it will have its own capability to
be tested here. However, we will want a way for platform code to request
to look like the actual CPU we're running on, especially if SoC devices
are being directly assigned.
David Gibson [Tue, 19 Apr 2011 01:54:52 +0000 (11:54 +1000)]
Place pseries vty devices at addresses more similar to existing machines
Currently the qemu pseries machine numbers its virtual serial devices
from 0. However, existing pSeries machines running pHyp number them from
0x30000000.
In theory these indices are arbitrary, since everything necessary for the
kernel to find them is advertised in the device tree. However the debian
installer, at least, incorrectly looks for a device named vty@30... to
determine whether to use the hypervisor console.
Therefore this patch moves the numbers we use to match the existing pHyp
practice, in order to workaround broken userspace apps of this type.
David Gibson [Tue, 19 Apr 2011 01:54:51 +0000 (11:54 +1000)]
Make pSeries 'model' property more closely resemble real hardware
Currently, the qemu emulated pseries machine puts
"qemu,emulated-pSeries-LPAR" in the device tree's root level 'model'
property. Unfortunately this confuses some installers and ybin, which
expect this to start with "IBM" on pSeries machines. This patch addresses
this problem, making the property more closely resemble the pattern of
existing real hardware.
Anton Blanchard [Tue, 19 Apr 2011 01:54:50 +0000 (11:54 +1000)]
pseries: Increase maximum CPUs to 256
The original pSeries machine was limited to 32 CPUs, more or less
arbitrarily. Particularly when we get SMT KVM guests it will be
pretty easy to exceed this. Therefore, raise the max number of CPUs
in a pseries machine guest to 256.
Gerd Hoffmann [Mon, 9 May 2011 07:44:03 +0000 (09:44 +0200)]
usb-musb: uninline functions
Prototype without "inline" keyword breaks the build with some gcc
versions. Noticed by Alexander Graf.
Fix this by removing the inline keywork everywhere. Some functions
can't be inlined anyway as the are referenced using function pointers.
Beside that gcc does a pretty good job on auto-inlining these days.
projects / qemu.git / log