sdk/emulator/qemu.git
12 years agoppc: Alter CPU state to mask out TCG unimplemented instructions as appropriate
David Gibson [Sun, 30 Oct 2011 15:51:24 +0000 (15:51 +0000)]
ppc: Alter CPU state to mask out TCG unimplemented instructions as appropriate

The CPU state contains two bitmaps, initialized from the CPU spec
which describes which instructions are implemented on the CPU.  A
couple of bits are defined which cover instructions (VSX and DFP)
which are not currently implemented in TCG.  So far, these are only
used to handle the case of -cpu host because a KVM guest can use
the instructions when the host CPU supports them.

However, it's a mild layering violation to simply not include those
bits in the CPU descriptions for those CPUs that do support them,
just because we can't handle them in TCG.  This patch corrects the
situation, so that the instruction bits _are_ shown correctly in the
cpu spec table, but are masked out from the cpu state in the non-KVM
case.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agopseries: Allow writes to KVM accelerated TCE table
David Gibson [Thu, 27 Oct 2011 15:56:31 +0000 (15:56 +0000)]
pseries: Allow writes to KVM accelerated TCE table

Sufficiently recent kernels include a KVM call to accelerate use of
PAPR TCE tables (IOMMU), which are used by PAPR virtual IO devices.
This involves qemu mapping the TCE table in from a kernel obtained fd,
which currently we do with PROT_READ only.  This is a hangover from
early (never released) versions of this kernel interface which only
permitted read-only mappings and required us to destroy and recreate
the table when we needed to clear it from qemu.

Now, the kernel permits read-write mappings, and we rely on this to
clear the table in spapr_vio_quiesce_one().  However, due to
insufficient testing, I forgot to update the actual mapping of the
table in kvmppc_create_spapr_tce() to add PROT_WRITE to the mmap().

This patch corrects the oversight.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agoKVM: PPC: Override host vmx/vsx/dfp only when information known
Alexander Graf [Mon, 24 Oct 2011 18:43:22 +0000 (20:43 +0200)]
KVM: PPC: Override host vmx/vsx/dfp only when information known

The -cpu host feature tries to find out the host capabilities based
on device tree information. However, we don't always have that available
because it's an optional property in dt.

So instead of force unsetting values depending on an unreliable source
of information, let's just try to be clever about it and not override
capabilities when we don't know the device tree pieces.

This fixes altivec with -cpu host on YDL PowerStations.

Reported-by: Nishanth Aravamudan <nacc@us.ibm.com>
Acked-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agoppc: Fix up usermode only builds
David Gibson [Sun, 23 Oct 2011 17:25:04 +0000 (17:25 +0000)]
ppc: Fix up usermode only builds

The recent usage of MemoryRegion in kvm_ppc.h breaks builds with
CONFIG_USER_ONLY=y.  This patch fixes it.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agopseries: Correct vmx/dfp handling in both KVM and TCG cases
David Gibson [Mon, 17 Oct 2011 18:15:41 +0000 (18:15 +0000)]
pseries: Correct vmx/dfp handling in both KVM and TCG cases

Currently, when KVM is enabled, the pseries machine checks if the host
CPU supports VMX, VSX and/or DFP instructions and advertises
accordingly in the guest device tree.  It does this regardless of what
CPU is selected on the command line.  On the other hand, when in TCG
mode, it never advertises any of these facilities, even basic VMX
(Altivec) which is supported in TCG.

Now that we have a -cpu host option for ppc, it is fairly
straightforward to fix both problems.  This patch changes the -cpu
host code to override the basic cpu spec derived from the PVR with
information queried from the host avout VMX, VSX and DFP capability.
The pseries code then uses the instruction availability advertised in
the cpu state to set the guest device tree correctly for both the KVM
and TCG cases.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agoPPC: Fail configure when libfdt is not available
Alexander Graf [Tue, 18 Oct 2011 00:10:36 +0000 (02:10 +0200)]
PPC: Fail configure when libfdt is not available

We have several targets in the PPC tree now that basically require libfdt
to function properly, namely the pseries and the e500 targets. This dependency
will rather increase than decrease in the future, so I want to make sure
that people building shiny new 1.0 actually have libfdt installed to get
rid of a few ifdefs in the code.

Warning: This patch will likely make configure fail for people who don't
select their own --target-list, but don't have libfdt development packages
installed. However, we really need this new dependency to move on.

Signed-off-by: Alexander Graf <agraf@suse.de>
---

v1 -> v2:

  - no paranthesis
  - no fdt check for config_pseries
  - add . in error message

12 years agoppc: Avoid decrementer related kvm exits
David Gibson [Sun, 16 Oct 2011 19:26:17 +0000 (19:26 +0000)]
ppc: Avoid decrementer related kvm exits

In __cpu_ppc_store_decr(), we set up a regular timer used to trigger
decrementer interrupts.  This is necessary to implement the decrementer
properly under TCG, but is unnecessary under KVM (true for both Book3S-PR
and Book3S-HV KVM variants), because the kernel handles generating and
delivering decrementer exceptions.

Under kvm, in fact, the timer causes expensive and unnecessary exits from
kvm to qemu.  This patch, therefore, disables setting the timer when kvm
is in use.

Signed-off-by: Anton Blanchard <anton@au1.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agoPPC: Disable non-440 CPUs for ppcemb target
Alexander Graf [Tue, 18 Oct 2011 00:00:51 +0000 (02:00 +0200)]
PPC: Disable non-440 CPUs for ppcemb target

The sole reason we have the ppcemb target is to support MMUs that have
less than the usual 4k possible page size. There are very few of these
chips and I don't want to add additional QA and testing burden to everyone
to ensure that code still works when TARGET_PAGE_SIZE is not 4k.

So this patch disables all CPUs except for MMU_BOOKE capable ones from
the ppcemb target.

Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agoPPC: Bump qemu-system-ppc to 64-bit physical address space
Alexander Graf [Mon, 17 Oct 2011 23:46:08 +0000 (01:46 +0200)]
PPC: Bump qemu-system-ppc to 64-bit physical address space

Some 32-bit PPC CPUs can use up to 36 bit of physical address space.
Treat them accordingly in the qemu-system-ppc binary type.

Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agopseries: Under kvm use guest cpu = host cpu by default
David Gibson [Wed, 12 Oct 2011 22:40:34 +0000 (22:40 +0000)]
pseries: Under kvm use guest cpu = host cpu by default

Now that we've implemented -cpu host for ppc, this patch updates the
pseries machine to use the host cpu as the guest cpu by default when
running under KVM.  This is important because under KVM Book3S-HV the guest
cpu _cannot_ be of a different type to the host cpu (at the moment
KVM Book3S-HV will silently virtualize the host cpu instead of whatever was
requested, but in future it is likely to simply refuse to run the VM if
a cpu model other than the host's is requested).

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agoppc: Add cpu defs for POWER7 revisions 2.1 and 2.3
David Gibson [Wed, 12 Oct 2011 22:40:33 +0000 (22:40 +0000)]
ppc: Add cpu defs for POWER7 revisions 2.1 and 2.3

This patch adds cpu specs to the table for POWER7 revisions 2.1 and 2.3.
This allows -cpu host to be used on these host cpus.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agoppc: First cut implementation of -cpu host
David Gibson [Wed, 12 Oct 2011 22:40:32 +0000 (22:40 +0000)]
ppc: First cut implementation of -cpu host

For convenience with kvm, x86 allows the user to specify -cpu host on the
qemu command line, which means make the guest cpu the same as the host
cpu.  This patch implements the same option for ppc targets.

For now, this just read the host PVR (Processor Version Register) and
selects one of our existing CPU specs based on it.  This means that the
option will not work if the host cpu is not supported by TCG, even if that
wouldn't matter for use under kvm.

In future, we can extend this in future to override parts of the cpu spec
based on information obtained from the host (via /proc/cpuinfo, the host
device tree, or explicit KVM calls).  That will let us handle cases where
the real kvm-virtualized CPU doesn't behave exactly like the TCG-emulated
CPU.  With appropriate annotation of the CPU specs we'll also then be able
to use host cpus under kvm even when there isn't a matching full TCG model.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agoppc: Remove broken partial PVR matching
David Gibson [Wed, 12 Oct 2011 22:40:31 +0000 (22:40 +0000)]
ppc: Remove broken partial PVR matching

The ppc target contains a ppc_find_by_pvr() function, which looks up a
CPU spec based on a PVR (that is, based on the value in the target cpu's
Processor Version Register).  PVR values contain information on both the
cpu model (upper 16 bits, usually) and on the precise revision (low 16
bits, usually).

ppc_find_by_pvr, as well as making exact PVR matches, attempts to find
"close" PVR matches, when we don't have a CPU spec for the exact revision
specified.  This sounds like a good idea, execpt that the current logic
is completely nonsensical.

It seems to assume CPU families are subdivided bit by bit in the PVR in a
way they just aren't.  Specifically, it requires a match on all bits of the
specified pvr up to the last non-zero bit.  This has the bizarre effect
that when the low bits are simply a sequential revision number (a common
though not universal pattern), then odd specified revisions must be matched
exactly, whereas even specified revisions will also match the next odd
revision, likewise for powers of 4, 8 and so forth.

To correctly do inexact matching we'd need to re-organize the table of CPU
specs to include a mask showing what PVR range the spec is compatible with
(similar to the cputable code in the Linux kernel).

For now, just remove the bogosity by only permitting exact PVR matches.
That at least makes the matching simple and consistent.  If we need inexact
matching we can add the necessary per-subfamily masks later.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agopseries: Update SLOF firmware image
David Gibson [Wed, 12 Oct 2011 19:10:30 +0000 (19:10 +0000)]
pseries: Update SLOF firmware image

This patch is a general update to the SLOF firmware image used on the
pseries machine.  This doesn't contain updates for specific features but
contains a number of bugfixes and enhancements in the main SLOF tree from
Thomas Huth.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agopseries: Add device tree properties for VMX/VSX and DFP under kvm
David Gibson [Mon, 10 Oct 2011 18:31:01 +0000 (18:31 +0000)]
pseries: Add device tree properties for VMX/VSX and DFP under kvm

Sufficiently recent PAPR specifications define properties "ibm,vmx"
and "ibm,dfp" on the CPU node which advertise whether the VMX vector
extensions (or the later VSX version) and/or the Decimal Floating
Point operations from IBM's recent POWER CPUs are available.

Currently we do not put these in the guest device tree and the guest
kernel will consequently assume they are not available.  This is good,
because they are not supported under TCG.  VMX is similar enough to
Altivec that it might be trivial to support, but VSX and DFP would
both require significant work to support in TCG.

However, when running under kvm on a host which supports these
instructions, there's no reason not to let the guest use them.  This
patch, therefore, checks for the relevant support on the host CPU
and, if present, advertises them to the guest as well.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agoppc: Generalize the kvmppc_get_clockfreq() function
David Gibson [Mon, 10 Oct 2011 18:31:00 +0000 (18:31 +0000)]
ppc: Generalize the kvmppc_get_clockfreq() function

Currently the kvmppc_get_clockfreq() function reads the host's clock
frequency from /proc/device-tree, which is useful to past to the guest
in KVM setups.  However, there are some other host properties
advertised in the device tree which can also be relevant to the
guests.

This patch, therefore, replaces kvmppc_get_clockfreq() which can
retrieve any named, single integer property from the host device
tree's CPU node.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agoSet an invalid-bits mask for each SPE instructions
Fabien Chouteau [Wed, 28 Sep 2011 05:54:05 +0000 (05:54 +0000)]
Set an invalid-bits mask for each SPE instructions

SPE instructions are defined by pairs. Currently, the invalid-bits mask is set
for the first instruction, but the second one can have a different mask.

example:
GEN_SPE(efdcmpeq,    efdcfs,      0x17, 0x0B, 0x00600000, 0x00180000, PPC_SPE_DOUBLE),

Signed-off-by: Fabien Chouteau <chouteau@adacore.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agopseries: Update SLOF firmware image
David Gibson [Thu, 29 Sep 2011 21:39:13 +0000 (21:39 +0000)]
pseries: Update SLOF firmware image

This patch updates the SLOF submodule and precompiled image.  The new
SLOF versions contains two changes of note:

 * The previous SLOF has a bug in SCSI condition handling that was
   exposed by recent updates to qemu's SCSI emulation.  This update
   fixes the bug.

 * The previous SLOF has a bug in its addressing of SCSI devices,
   which can be exposed under certain conditions.  The new SLOF also
   fixes this.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agopseries: Use Book3S-HV TCE acceleration capabilities
David Gibson [Thu, 29 Sep 2011 21:39:12 +0000 (21:39 +0000)]
pseries: Use Book3S-HV TCE acceleration capabilities

The pseries machine of qemu implements the TCE mechanism used as a
virtual IOMMU for the PAPR defined virtual IO devices.  Because the
PAPR spec only defines a small DMA address space, the guest VIO
drivers need to update TCE mappings very frequently - the virtual
network device is particularly bad.  This means many slow exits to
qemu to emulate the H_PUT_TCE hypercall.

Sufficiently recent kernels allow this to be mitigated by implementing
H_PUT_TCE in the host kernel.  To make use of this, however, qemu
needs to initialize the necessary TCE tables, and map them into itself
so that the VIO device implementations can retrieve the mappings when
they access guest memory (which is treated as a virtual DMA
operation).

This patch adds the necessary calls to use the KVM TCE acceleration.
If the kernel does not support acceleration, or there is some other
error creating the accelerated TCE table, then it will still fall back
to full userspace TCE implementation.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agopseries: Allow KVM Book3S-HV on PPC970 CPUS
David Gibson [Thu, 29 Sep 2011 21:39:11 +0000 (21:39 +0000)]
pseries: Allow KVM Book3S-HV on PPC970 CPUS

At present, using the hypervisor aware Book3S-HV KVM will only work
with qemu on POWER7 CPUs.  PPC970 CPUs also have hypervisor
capability, but they lack the VRMA feature which makes assigning guest
memory easier.

In order to allow KVM Book3S-HV on PPC970, we need to specially
allocate the first chunk of guest memory (the "Real Mode Area" or
RMA), so that it is physically contiguous.

Sufficiently recent host kernels allow such contiguous RMAs to be
allocated, with a kvm capability advertising whether the feature is
available and/or necessary on this hardware.  This patch enables qemu
to use this support, thus allowing kvm acceleration of pseries qemu
machines on PPC970 hardware.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
---

agraf: fix to use memory api

12 years agopseries: Support SMT systems for KVM Book3S-HV
David Gibson [Thu, 29 Sep 2011 21:39:10 +0000 (21:39 +0000)]
pseries: Support SMT systems for KVM Book3S-HV

Alex Graf has already made qemu support KVM for the pseries machine
when using the Book3S-PR KVM variant (which runs the guest in
usermode, emulating supervisor operations).  This code allows gets us
very close to also working with KVM Book3S-HV (using the hypervisor
capabilities of recent POWER CPUs).

This patch moves us another step towards Book3S-HV support by
correctly handling SMT (multithreaded) POWER CPUs.  There are two
parts to this:

 * Querying KVM to check SMT capability, and if present, adjusting the
   cpu numbers that qemu assigns to cause KVM to assign guest threads
   to cores in the right way (this isn't automatic, because the POWER
   HV support has a limitation that different threads on a single core
   cannot be in different guests at the same time).

 * Correctly informing the guest OS of the SMT thread to core mappings
   via the device tree.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agoppc/e500_pci: Fix an array overflow issue
Liu Yu-B13201 [Thu, 29 Sep 2011 17:52:50 +0000 (17:52 +0000)]
ppc/e500_pci: Fix an array overflow issue

When access PPCE500_PCI_IW1 the previous index get overflow.
The patch fix the issue and update all to keep consistent style.

Signed-off-by: Liu Yu <yu.liu@freescale.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agoppc/e500_pci: Fix code style
Liu Yu-B13201 [Thu, 29 Sep 2011 17:52:49 +0000 (17:52 +0000)]
ppc/e500_pci: Fix code style

Put trailing statements on next line.

Signed-off-by: Liu Yu <yu.liu@freescale.com>
Reviewed-by: Andreas Färber <andreas.faerber@web.de>
Signed-off-by: Alexander Graf <agraf@suse.de>
12 years agoMAINTAINERS: update wiki URL and machine names for target-xtensa
Max Filippov [Thu, 27 Oct 2011 22:11:05 +0000 (02:11 +0400)]
MAINTAINERS: update wiki URL and machine names for target-xtensa

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agotcg: Optimize some forms of deposit.
Richard Henderson [Thu, 27 Oct 2011 21:15:00 +0000 (14:15 -0700)]
tcg: Optimize some forms of deposit.

If the deposit replaces the entire word, optimize to a move.

If we're inserting to the top of the word, avoid the mask of arg2
as we'll be shifting out all of the garbage and shifting in zeros.

If the host is 32-bit, reduce a 64-bit deposit to a 32-bit deposit
when possible.

Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agohw/9pfs: Make VirtFS tracing work correctly
Aneesh Kumar K.V [Mon, 24 Oct 2011 09:39:49 +0000 (15:09 +0530)]
hw/9pfs: Make VirtFS tracing work correctly

this patch fix multiple issues with VirtFS tracing.
a) Add tracepoint to the correct code path. We handle error in complete_pdu
b) Fix indentation in python script
c) Fix variable naming issue in python script

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoexec-all: Fix void pointer arithmetic
Stefan Weil [Sun, 23 Oct 2011 06:19:10 +0000 (08:19 +0200)]
exec-all: Fix void pointer arithmetic

Adding an offset to a void pointer works with gcc but is not allowed
by the current C standards. With -pedantic, gcc complains:

exec-all.h:344: error: pointer of type ‘void *’ used in arithmetic

Fix this, and also replace (unsigned long) by (uintptr_t) in the same
statement.

Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoAdd linux-headers/asm to .gitignore
David Gibson [Thu, 20 Oct 2011 00:41:01 +0000 (11:41 +1100)]
Add linux-headers/asm to .gitignore

linux-headers/asm is a symlink generated during configure.  It should not,
therefore be committed to git, nor show up in git diffs and the like.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoMerge branch 'rth/vis2' of git://repo.or.cz/qemu/rth
Blue Swirl [Thu, 27 Oct 2011 20:27:07 +0000 (20:27 +0000)]
Merge branch 'rth/vis2' of git://repo.or.cz/qemu/rth

* 'rth/vis2' of git://repo.or.cz/qemu/rth:
  target-sparc: Implement FALIGNDATA inline.
  target-sparc: Implement BMASK/BSHUFFLE.
  target-sparc: Implement ALIGNADDR* inline.
  target-sparc: Implement EDGE* instructions.
  target-sparc: Implement fpack{16,32,fix}.
  target-sparc: Implement PDIST.
  target-sparc: Do exceptions management fully inside the helpers.
  target-sparc: Change fpr representation to doubles.
  target-sparc: Undo cpu_fpr rename.
  target-sparc: Extract float128 move to a function.
  target-sparc: Extract common code for floating-point operations.
  target-sparc: Make FPU/VIS helpers const when possible.
  target-sparc: Pass float64 parameters instead of dt0/1 temporaries.
  target-sparc: Add accessors for double-precision fpr access.
  target-sparc: Mark fprs dirty in store accessor.
  target-sparc: Add accessors for single-precision fpr access.

12 years agoMerge branch 'target-arm.for-upstream' of git://git.linaro.org/people/pmaydell/qemu-arm
Andrzej Zaborowski [Wed, 26 Oct 2011 23:02:46 +0000 (01:02 +0200)]
Merge branch 'target-arm.for-upstream' of git://git.linaro.org/people/pmaydell/qemu-arm

12 years agotarget-sparc: Implement FALIGNDATA inline.
Richard Henderson [Tue, 18 Oct 2011 16:41:30 +0000 (09:41 -0700)]
target-sparc: Implement FALIGNDATA inline.

This is a relatively simple sequence of shifts.

Signed-off-by: Richard Henderson <rth@twiddle.net>
12 years agotarget-sparc: Implement BMASK/BSHUFFLE.
Richard Henderson [Tue, 18 Oct 2011 16:24:43 +0000 (09:24 -0700)]
target-sparc: Implement BMASK/BSHUFFLE.

Signed-off-by: Richard Henderson <rth@twiddle.net>
12 years agotarget-sparc: Implement ALIGNADDR* inline.
Richard Henderson [Tue, 18 Oct 2011 15:35:58 +0000 (08:35 -0700)]
target-sparc: Implement ALIGNADDR* inline.

While ALIGNADDR was implemented out-of-line, ALIGNADDRL was not
implemeneted at all.  However, this is a very simple operation
so we're better off doing this inline.

Signed-off-by: Richard Henderson <rth@twiddle.net>
12 years agotarget-sparc: Implement EDGE* instructions.
Richard Henderson [Tue, 18 Oct 2011 02:57:23 +0000 (19:57 -0700)]
target-sparc: Implement EDGE* instructions.

Signed-off-by: Richard Henderson <rth@twiddle.net>
12 years agotarget-sparc: Implement fpack{16,32,fix}.
Richard Henderson [Tue, 18 Oct 2011 01:03:47 +0000 (18:03 -0700)]
target-sparc: Implement fpack{16,32,fix}.

Signed-off-by: Richard Henderson <rth@twiddle.net>
12 years agotarget-sparc: Implement PDIST.
Richard Henderson [Tue, 18 Oct 2011 00:32:26 +0000 (17:32 -0700)]
target-sparc: Implement PDIST.

Signed-off-by: Richard Henderson <rth@twiddle.net>
12 years agotarget-sparc: Do exceptions management fully inside the helpers.
Richard Henderson [Mon, 17 Oct 2011 18:25:56 +0000 (11:25 -0700)]
target-sparc: Do exceptions management fully inside the helpers.

This reduces the size of the individual translation blocks, since
we only emit a single call for each FOP rather than three.  In
addition, clear_float_exceptions expands inline to a single byte store.

Signed-off-by: Richard Henderson <rth@twiddle.net>
12 years agotarget-sparc: Change fpr representation to doubles.
Richard Henderson [Mon, 17 Oct 2011 17:42:49 +0000 (10:42 -0700)]
target-sparc: Change fpr representation to doubles.

This allows a more efficient representation for 64-bit hosts.
It should be about the same for 32-bit hosts, as we can still
access the individual pieces of the double.

Signed-off-by: Richard Henderson <rth@twiddle.net>
12 years agotarget-sparc: Undo cpu_fpr rename.
Richard Henderson [Sat, 15 Oct 2011 19:12:02 +0000 (12:12 -0700)]
target-sparc: Undo cpu_fpr rename.

Signed-off-by: Richard Henderson <rth@twiddle.net>
12 years agotarget-sparc: Extract float128 move to a function.
Richard Henderson [Sat, 15 Oct 2011 19:00:08 +0000 (12:00 -0700)]
target-sparc: Extract float128 move to a function.

Signed-off-by: Richard Henderson <rth@twiddle.net>
12 years agotarget-sparc: Extract common code for floating-point operations.
Richard Henderson [Sat, 15 Oct 2011 18:52:00 +0000 (11:52 -0700)]
target-sparc: Extract common code for floating-point operations.

Signed-off-by: Richard Henderson <rth@twiddle.net>
12 years agotarget-sparc: Make FPU/VIS helpers const when possible.
Richard Henderson [Wed, 19 Oct 2011 21:56:43 +0000 (14:56 -0700)]
target-sparc: Make FPU/VIS helpers const when possible.

This also removes the unused ENV parameter from these helpers.

Signed-off-by: Richard Henderson <rth@twiddle.net>
12 years agotarget-sparc: Pass float64 parameters instead of dt0/1 temporaries.
Richard Henderson [Sat, 15 Oct 2011 17:20:20 +0000 (10:20 -0700)]
target-sparc: Pass float64 parameters instead of dt0/1 temporaries.

Signed-off-by: Richard Henderson <rth@twiddle.net>
12 years agotarget-sparc: Add accessors for double-precision fpr access.
Richard Henderson [Fri, 14 Oct 2011 22:47:35 +0000 (15:47 -0700)]
target-sparc: Add accessors for double-precision fpr access.

Begin using i64 quantities to manipulate double-precision values.
On a 64-bit host this will, for the moment, generate less efficient
code; on a 32-bit host code quality should be largely unchanged.
Code quality for 64-bit will be adjusted with a subsequent patch.

Signed-off-by: Richard Henderson <rth@twiddle.net>
12 years agotarget-sparc: Mark fprs dirty in store accessor.
Richard Henderson [Fri, 14 Oct 2011 22:03:25 +0000 (15:03 -0700)]
target-sparc: Mark fprs dirty in store accessor.

Signed-off-by: Richard Henderson <rth@twiddle.net>
12 years agotarget-sparc: Add accessors for single-precision fpr access.
Richard Henderson [Fri, 14 Oct 2011 21:58:32 +0000 (14:58 -0700)]
target-sparc: Add accessors for single-precision fpr access.

Load, store, and "create destination".  This version attempts to
change the behaviour of the translator as little as possible.  We
previously used cpu_tmp32 as the temporary destination, and we
continue to use that.  This will eventually allow a change in
representation of the fprs.

Change the name of the cpu_fpr array to make certain that all
instances are converted.

Signed-off-by: Richard Henderson <rth@twiddle.net>
12 years agoSparc: split load and store op helpers
Blue Swirl [Mon, 1 Aug 2011 10:15:51 +0000 (10:15 +0000)]
Sparc: split load and store op helpers

Move load and store op helpers top ldst_helper.c.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoSparc: convert win_helper to trace framework
Blue Swirl [Sun, 11 Sep 2011 15:53:35 +0000 (15:53 +0000)]
Sparc: convert win_helper to trace framework

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoSparc: convert interrupt helpers to trace framework
Blue Swirl [Sun, 11 Sep 2011 15:05:41 +0000 (15:05 +0000)]
Sparc: convert interrupt helpers to trace framework

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoSparc: convert mmu_helper to trace framework
Blue Swirl [Sun, 11 Sep 2011 14:51:24 +0000 (14:51 +0000)]
Sparc: convert mmu_helper to trace framework

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoSparc: split MMU helpers
Blue Swirl [Sun, 11 Sep 2011 11:30:01 +0000 (11:30 +0000)]
Sparc: split MMU helpers

Move MMU helpers to mmu_helper.c.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoSparc: fix coding style in helper.c
Blue Swirl [Sun, 11 Sep 2011 12:03:08 +0000 (12:03 +0000)]
Sparc: fix coding style in helper.c

Before the next patch, fix coding style of the areas affected.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoSparc: avoid AREG0 for division op helpers
Blue Swirl [Mon, 4 Jul 2011 18:15:42 +0000 (18:15 +0000)]
Sparc: avoid AREG0 for division op helpers

Make [su]div{,cc} helpers take a parameter for CPUState instead
of relying on global env. Move the functions to helper.c.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoSparc: avoid AREG0 for softint op helpers and Leon cache control
Blue Swirl [Mon, 1 Aug 2011 09:20:58 +0000 (09:20 +0000)]
Sparc: avoid AREG0 for softint op helpers and Leon cache control

Make softint op helpers and Leon cache irq manager take a parameter
for CPUState instead of relying on global env. Move the functions
to int{32,64}_helper.c.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoSparc: avoid AREG0 for CWP and PSTATE helpers
Blue Swirl [Sun, 3 Jul 2011 21:01:59 +0000 (21:01 +0000)]
Sparc: avoid AREG0 for CWP and PSTATE helpers

Make CWP and PSTATE helpers take a parameter for CPUState instead
of relying on global env. Remove wrapper functions.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoMerge remote-tracking branch 'kwolf/for-anthony' into staging
Anthony Liguori [Wed, 26 Oct 2011 15:29:24 +0000 (10:29 -0500)]
Merge remote-tracking branch 'kwolf/for-anthony' into staging

12 years agomain-loop: Add missing include file
Stefan Weil [Tue, 25 Oct 2011 20:23:17 +0000 (22:23 +0200)]
main-loop: Add missing include file

stdint.h defines the POSIX data types and is needed
for MinGW-w64 (and maybe other hosts).

v2: Instead of adding stdint.h directly, qemu-common.h is now
included and duplicate include statements were removed.

Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
12 years agotarget-sparc: Fix use of g_new0 / g_free
Stefan Weil [Tue, 25 Oct 2011 05:16:25 +0000 (07:16 +0200)]
target-sparc: Fix use of g_new0 / g_free

g_malloc0 needs g_free instead of free.
While fixing this, I also replaced g_malloc0 by g_new0
as was suggested by Stuart Brady.

Cc: Blue Swirl <blauwirbel@gmail.com>
Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agotarget-sparc: Fix order of function parameters
Stefan Weil [Mon, 24 Oct 2011 20:29:48 +0000 (22:29 +0200)]
target-sparc: Fix order of function parameters

The MinGW-w64 gcc complains about wrong parameters for
gen_helper_fpadd16_s and three other functions.

gen_helper_fpadd16_s is declared like this (hidden in lots of macros):

static inline void
 gen_helper_fpadd16s(TCGv_i32 retval, TCGv_ptr arg1,
                     TCGv_i32 arg2, TCGv_i32 arg3);

So it looks like cpu_env should be the 2nd parameter.

Please review this patch as I have no environment to test it
(maybe the 1st parameter should be cpu_dst?).

Cc: Blue Swirl <blauwirbel@gmail.com>
Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agohda: do not mix output and input stream states, RHBZ #740493
Marc-André Lureau [Tue, 25 Oct 2011 14:53:01 +0000 (16:53 +0200)]
hda: do not mix output and input stream states, RHBZ #740493

Windows 7 may use the same stream number for input and output.
Current code will confuse streams.

Changes since v1:
- keep running_compat[] for migration version 1
- add running_real[] for migration version 2

Signed-off-by: Marc-Andr? Lureau <marcandre.lureau@redhat.com>
Signed-off-by: malc <av1474@comtv.ru>
12 years agohda: do not mix output and input streams, RHBZ #740493
Marc-André Lureau [Tue, 25 Oct 2011 14:53:00 +0000 (16:53 +0200)]
hda: do not mix output and input streams, RHBZ #740493

Windows 7 may use the same stream number for input and output.
That will result in lot of garbage on playback.

The hardcoded value of 4 needs to be in sync with GCAP streams
description and IN/OUT registers.

Signed-off-by: Marc-Andr? Lureau <marcandre.lureau@redhat.com>
Signed-off-by: malc <av1474@comtv.ru>
12 years agoMerge remote-tracking branch 'bonzini/split-main-loop-for-anthony' into staging
Anthony Liguori [Mon, 24 Oct 2011 15:51:12 +0000 (10:51 -0500)]
Merge remote-tracking branch 'bonzini/split-main-loop-for-anthony' into staging

12 years agoAdd stdio char device on windows
Fabien Chouteau [Thu, 6 Oct 2011 14:37:51 +0000 (16:37 +0200)]
Add stdio char device on windows

Simple implementation of an stdio char device on Windows.

Signed-off-by: Fabien Chouteau <chouteau@adacore.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoSparc: split CWP and PSTATE op helpers
Blue Swirl [Mon, 1 Aug 2011 09:03:20 +0000 (09:03 +0000)]
Sparc: split CWP and PSTATE op helpers

Move CWP and PSTATE op helpers to win_helper.c.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoSparc: avoid AREG0 for lazy condition code helpers
Blue Swirl [Sun, 3 Jul 2011 16:01:57 +0000 (16:01 +0000)]
Sparc: avoid AREG0 for lazy condition code helpers

Make lazy condition code helpers take a parameter for CPUState instead
of relying on global env.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoSparc: split lazy condition code handling op helpers
Blue Swirl [Mon, 1 Aug 2011 08:37:36 +0000 (08:37 +0000)]
Sparc: split lazy condition code handling op helpers

Move lazy condition code handling op helpers to cc_helper.c.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoSparc: avoid AREG0 for float and VIS ops
Blue Swirl [Sun, 3 Jul 2011 10:42:23 +0000 (10:42 +0000)]
Sparc: avoid AREG0 for float and VIS ops

Make floating point and VIS ops take a parameter for CPUState instead
of relying on global env.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoSparc: split FPU and VIS op helpers
Blue Swirl [Mon, 1 Aug 2011 07:37:45 +0000 (07:37 +0000)]
Sparc: split FPU and VIS op helpers

Move FPU op helpers to fop_helper.c. Move VIS op helpers to vis_helper.c,
compile it only for Sparc64.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoSparc: fix coding style
Blue Swirl [Sun, 3 Jul 2011 15:22:03 +0000 (15:22 +0000)]
Sparc: fix coding style

Before the next patches, fix coding style of the areas affected.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoSparc: avoid AREG0 for raise_exception and helper_debug
Blue Swirl [Sun, 3 Jul 2011 08:19:42 +0000 (08:19 +0000)]
Sparc: avoid AREG0 for raise_exception and helper_debug

Make raise_exception() and helper_debug() take a parameter for
CPUState instead of relying on global env. Move the functions
to helper.c.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoSparc: move trivial functions from op_helper.c
Blue Swirl [Sun, 3 Jul 2011 07:05:50 +0000 (07:05 +0000)]
Sparc: move trivial functions from op_helper.c

These functions don't need access to CPUState or already pass it,
so relocating them from op_helper.c to helper.c and int64_helper.c
is trivial.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoSparc: split helper.c
Blue Swirl [Sun, 11 Sep 2011 09:33:40 +0000 (09:33 +0000)]
Sparc: split helper.c

Move CPU init to cpu_init.c and interrupt handling to int32_helper.c
for Sparc32 and int64_helper.c for Sparc64.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agoSparc: fix coding style
Blue Swirl [Sun, 11 Sep 2011 09:42:13 +0000 (09:42 +0000)]
Sparc: fix coding style

Before the next patch, fix coding style of the areas affected.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agom48t59: fix write access
Blue Swirl [Sat, 15 Oct 2011 08:05:18 +0000 (08:05 +0000)]
m48t59: fix write access

Fix incorrect order of arguments, letting writes to NVRAM succeed.

It looks like guests never write to the device, only read from it, since the bug
originates back to 819385c58b319d9f80d676cefaed0610118f03ac.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agom48t59: drop obsolete address base arithmetic
Blue Swirl [Sat, 15 Oct 2011 07:57:49 +0000 (07:57 +0000)]
m48t59: drop obsolete address base arithmetic

Remove now incorrect address base arithmetic, missed by
9936d6e42392f1440505dfa9df065eabd251cadf. Fixes Sparc64 boot.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agopci_bridge: fix typo
Blue Swirl [Sun, 16 Oct 2011 14:43:00 +0000 (14:43 +0000)]
pci_bridge: fix typo

Reviewed-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
12 years agosimplify main loop functions
Paolo Bonzini [Mon, 12 Sep 2011 12:03:13 +0000 (14:03 +0200)]
simplify main loop functions

Provide a clean example of how to use the main loop in the tools.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agoRevert to a hand-made select loop
Paolo Bonzini [Mon, 12 Sep 2011 12:59:42 +0000 (14:59 +0200)]
Revert to a hand-made select loop

This reverts commit c82dc29a9112f34e0a51cad9a412cf6d9d05dfb2
and 4d88a2ac8643265108ef1fb47ceee5d7b28e19f2.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agomain-loop: create main-loop.c
Paolo Bonzini [Tue, 13 Sep 2011 08:30:52 +0000 (10:30 +0200)]
main-loop: create main-loop.c

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agomain-loop: create main-loop.h
Paolo Bonzini [Mon, 12 Sep 2011 14:44:30 +0000 (16:44 +0200)]
main-loop: create main-loop.h

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agoqemu-timer: do not use RunState change handlers
Paolo Bonzini [Tue, 27 Sep 2011 16:23:14 +0000 (18:23 +0200)]
qemu-timer: do not use RunState change handlers

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agoqemu-timer: move more stuff out of qemu-timer.c
Paolo Bonzini [Mon, 12 Sep 2011 14:21:44 +0000 (16:21 +0200)]
qemu-timer: move more stuff out of qemu-timer.c

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agoqemu-timer: use atexit for quit_timers
Paolo Bonzini [Mon, 19 Sep 2011 08:18:51 +0000 (10:18 +0200)]
qemu-timer: use atexit for quit_timers

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agoqemu-timer: do not refer to runstate_is_running()
Paolo Bonzini [Mon, 12 Sep 2011 12:40:36 +0000 (14:40 +0200)]
qemu-timer: do not refer to runstate_is_running()

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agoqemu-timer: move icount to cpus.c
Paolo Bonzini [Mon, 12 Sep 2011 11:57:37 +0000 (13:57 +0200)]
qemu-timer: move icount to cpus.c

None of this is needed by tools, and most of it can even be made static
inside cpus.c.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agoqemu-timer: more clock functions
Paolo Bonzini [Mon, 12 Sep 2011 13:50:16 +0000 (15:50 +0200)]
qemu-timer: more clock functions

These will be used when moving icount accounting to cpus.c.

Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agoqemu-timer: move common code to qemu_rearm_alarm_timer
Paolo Bonzini [Mon, 14 Mar 2011 08:45:38 +0000 (09:45 +0100)]
qemu-timer: move common code to qemu_rearm_alarm_timer

Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agoqemu-timer: remove active_timers array
Paolo Bonzini [Tue, 13 Sep 2011 09:42:26 +0000 (11:42 +0200)]
qemu-timer: remove active_timers array

Embed the list in the QEMUClock instead.

Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agoremove unused function
Paolo Bonzini [Mon, 12 Sep 2011 13:17:57 +0000 (15:17 +0200)]
remove unused function

Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
12 years agocompatfd.c: Don't pass NULL pointer to SYS_signalfd
Peter Maydell [Thu, 13 Oct 2011 17:45:37 +0000 (18:45 +0100)]
compatfd.c: Don't pass NULL pointer to SYS_signalfd

Don't pass a NULL pointer in to SYS_signalfd in qemu_signalfd_available():
this isn't valid and Valgrind complains about it.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Andrzej Zaborowski <andrew.zaborowski@intel.com>
12 years agolinux-user: Fix broken "-version" option
Peter Maydell [Thu, 29 Sep 2011 14:48:12 +0000 (15:48 +0100)]
linux-user: Fix broken "-version" option

Fix the "-version" option, which was accidentally broken in commit
fc9c541:
 * exit after printing version information rather than proceeding
   blithely onward (and likely printing the full usage message)
 * correct the cut-n-paste error in the usage message for it
 * don't insist on the presence of a following argument for
   options which don't take an argument (this was preventing
   'qemu-arm -version' from working)
 * remove a spurious argc check from the beginning of main() which
   meant 'QEMU_VERSION=1 qemu-arm' didn't work.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Andrzej Zaborowski <andrew.zaborowski@intel.com>
12 years agoblock: change discard to co_discard
Paolo Bonzini [Thu, 20 Oct 2011 11:16:25 +0000 (13:16 +0200)]
block: change discard to co_discard

Since coroutine operation is now mandatory, convert both bdrv_discard
implementations to coroutines.  For qcow2, this means taking the lock
around the operation.  raw-posix remains synchronous.

The bdrv_discard callback is then unused and can be eliminated.

Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agoblock: change flush to co_flush
Paolo Bonzini [Thu, 20 Oct 2011 11:16:24 +0000 (13:16 +0200)]
block: change flush to co_flush

Since coroutine operation is now mandatory, convert all bdrv_flush
implementations to coroutines.  For qcow2, this means taking the lock.
Other implementations are simpler and just forward bdrv_flush to the
underlying protocol, so they can avoid the lock.

The bdrv_flush callback is then unused and can be eliminated.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agoblock: take lock around bdrv_write implementations
Paolo Bonzini [Thu, 20 Oct 2011 11:16:23 +0000 (13:16 +0200)]
block: take lock around bdrv_write implementations

This does the first part of the conversion to coroutines, by
wrapping bdrv_write implementations to take the mutex.

Drivers that implement bdrv_write rather than bdrv_co_writev can
then benefit from asynchronous operation (at least if the underlying
protocol supports it, which is not the case for raw-win32), even
though they still operate with a bounce buffer.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agoblock: take lock around bdrv_read implementations
Paolo Bonzini [Thu, 20 Oct 2011 11:16:22 +0000 (13:16 +0200)]
block: take lock around bdrv_read implementations

This does the first part of the conversion to coroutines, by
wrapping bdrv_read implementations to take the mutex.

Drivers that implement bdrv_read rather than bdrv_co_readv can
then benefit from asynchronous operation (at least if the underlying
protocol supports it, which is not the case for raw-win32), even
though they still operate with a bounce buffer.

raw-win32 does not need the lock, because it cannot yield.
nbd also doesn't probably, but better be safe.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agoblock: add a CoMutex to synchronous read drivers
Paolo Bonzini [Thu, 20 Oct 2011 11:16:21 +0000 (13:16 +0200)]
block: add a CoMutex to synchronous read drivers

The big conversion of bdrv_read/write to coroutines caused the two
homonymous callbacks in BlockDriver to become reentrant.  It goes
like this:

1) bdrv_read is now called in a coroutine, and calls bdrv_read or
bdrv_pread.

2) the nested bdrv_read goes through the fast path in bdrv_rw_co_entry;

3) in the common case when the protocol is file, bdrv_co_do_readv calls
bdrv_co_readv_em (and from here goes to bdrv_co_io_em), which yields
until the AIO operation is complete;

4) if bdrv_read had been called from a bottom half, the main loop
is free to iterate again: a device model or another bottom half
can then come and call bdrv_read again.

This applies to all four of read/write/flush/discard.  It would also
apply to is_allocated, but it is not used from within coroutines:
besides qemu-img.c and qemu-io.c, which operate synchronously, the
only user is the monitor.  Copy-on-read will introduce a use in the
block layer, and will require converting it.

The solution is "simply" to convert all drivers to coroutines!  We
just need to add a CoMutex that is taken around affected operations.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agovmdk: clean up open
Paolo Bonzini [Thu, 20 Oct 2011 11:16:20 +0000 (13:16 +0200)]
vmdk: clean up open

Move vmdk_parent_open to vmdk_open.  There's another path how
vmdk_parent_open can be reached:

  vmdk_parse_extents() ->  vmdk_open_sparse() ->  vmdk_open_vmdk4() ->
  vmdk_open_desc_file().

If that can happen, however, the code is bogus.  vmdk_parent_open
reads from bs->file:

    if (bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE) != DESC_SIZE) {

but it is always called with s->desc_offset == 0 and with the same
bs->file.  So the data that vmdk_parent_open reads comes always from the
same place, and anyway there is only one place where it can write it,
namely bs->backing_file.

So, if it cannot happen, the patched code is okay.

It is also possible that the recursive call can happen, but only once.  In
that case there would still be a bug in vmdk_open_desc_file setting
s->desc_offset = 0, but the patched code is okay.

Finally, in the case where multiple recursive calls can happen the code
would need to be rewritten anyway.  It is likely that this would anyway
involve adding several parameters to vmdk_parent_open, and calling it from
vmdk_open_vmdk4.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agovmdk: fix return values of vmdk_parent_open
Paolo Bonzini [Thu, 20 Oct 2011 11:16:19 +0000 (13:16 +0200)]
vmdk: fix return values of vmdk_parent_open

While vmdk_open_desc_file (touched by the patch) correctly changed -1
to -EINVAL, vmdk_open did not.  Fix it directly in vmdk_parent_open.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12 years agopc: Fix floppy drives with if=none
Kevin Wolf [Thu, 20 Oct 2011 14:37:26 +0000 (16:37 +0200)]
pc: Fix floppy drives with if=none

Commit 63ffb564 broke floppy devices specified on the command line like
-drive file=...,if=none,id=floppy -global isa-fdc.driveA=floppy because it
relies on drive_get() which works only with -fda/-drive if=floppy.

This patch resembles what we're already doing for IDE, i.e. remember the floppy
device that was created and use that to extract the BlockDriverStates where
needed.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
12 years agoqcow2: Fix bdrv_write_compressed error handling
Kevin Wolf [Tue, 18 Oct 2011 15:12:44 +0000 (17:12 +0200)]
qcow2: Fix bdrv_write_compressed error handling

If during allocation of compressed clusters the cluster was already allocated
uncompressed, fail and properly release the l2_table (the latter avoids a
failed assertion).

While at it, make it return some real error numbers instead of -1.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Dong Xu Wang <wdongxu@linux.vnet.ibm.com>