pseries: Rewrite CAS PVR compatibility logic

During boot, PAPR guests negotiate CPU model support with the
ibm,client-architecture-support mechanism.  The logic to implement this in
qemu is very convoluted.  This cleans it up to be cleaner, using the new
ppc_check_compat() call.

The new logic for choosing a compatibility mode is:
    1. Usually, use the most recent compatibility mode that is
            a) supported by the guest
            b) supported by the CPU
        and c) no later than the maximum allowed (if specified)
    2. If no suitable compatibility mode was found, the guest *does*
       support this CPU explicitly, and no maximum compatibility mode is
       specified, then use "raw" mode for the current CPU
    3. Otherwise, fail the boot.

This differs from the results of the old code: the old code preferred using
"raw" mode to a compatibility mode, whereas the new code prefers a
compatibility mode if available.  Using compatibility mode preferentially
means that we're more likely to be able to migrate the guest to a similar
but not identical host.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

pxb: Restrict to x86

The PCI Expander Bridge (PXB) device is essentially a hack to allow
different PCIe devices to be assigned to different NUMA nodes on x86.  Each
PXB is sort-of a separate PCI host bridge, except that its config space
is shared with the config space of the main PCI host bridge, rather than
being independent.

This is only necessary if the platform doesn't (easily) allow truly
independent PCI host bridges.  AFAIK that's just x86.

This patch makes it possible to configure PXB out of the build, and adjusts
the default configs so it's only included on x86 targets.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Michael S. Tsirkin <mst@redhat.com>

target-ppc: Add xsxsigqp instructions

xsxsigqp: VSX Scalar Extract Significand Quad Precision

Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Add xsxsigdp instruction

xsxsigdp: VSX Scalar Extract Significand Dual Precision

Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Add xsxexpqp instruction

xsxexpqp: VSX Scalar Extract Exponent Quad Precision

Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Add xsxexpdp instruction

xsxexpdp: VSX Scalar Extract Exponent Dual Precision

Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Use correct precision for FPRF setting

Use correct FP precision when setting FPRF in FP conversion helpers
instead of always assuming float64 precision.

Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Add xscvdphp, xscvhpdp

xscvdphp: VSX Scalar round & Convert Double-Precision format to
          Half-Precision format
xscvhpdp: VSX Scalar Convert Half-Precision format to
          Double-Precision format

Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Rename helper_compute_fprf to helper_compute_fprf_float64

Since helper_compute_fprf() works on float64 argument, rename it
to helper_compute_fprf_float64(). Also use a macro to generate
helper_compute_fprf_float64() so that float128 version of the same
helper can be introduced easily later.

Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Replace isden by float64_is_zero_or_denormal

Replace isden() by float64_is_zero_or_denormal() so that code in
helper_compute_fprf() can be reused to work with float128 argument.

Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Use float64 arg in helper_compute_fprf()

Use float64 argument instead of unit64_t in helper_compute_fprf()
This allows code in helper_compute_fprf() to be reused later to
work with float128 argument too.

Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

prep: add IBM RS/6000 7020 (40p) machine emulation

Machine supports both Open Hack'Ware and OpenBIOS.
Open Hack'Ware is the default because OpenBIOS is currently unable to boot
PReP boot partitions or PReP kernels.

Signed-off-by: Hervé Poussineau <hpoussin@reactos.org>
[dwg: Correct compile failure with KVM located by Thomas Huth]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

prep: add IBM RS/6000 7020 (40p) memory controller

Signed-off-by: Hervé Poussineau <hpoussin@reactos.org>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
[dwg: Added CONFIG_RS6000_MC to ppc64 or it breaks testcases]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

prep: add PReP System I/O

This device is a partial duplicate of System I/O device available in hw/ppc/prep.c
This new one doesn't have all the Motorola-specific registers.
The old one should be deprecated and removed with the 'prep' machine.

Partial documentation available at
ftp://ftp.software.ibm.com/rs6000/technology/spec/srp1_1.exe
section 6.1.5 (I/O Device Mapping)

Signed-off-by: Hervé Poussineau <hpoussin@reactos.org>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Add xxinsertw instruction

xxinsertw: VSX Vector Insert Word

Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Add xxextractuw instruction

xxextractuw: VSX Vector Extract Unsigned Word

Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

hw/ppc: QOM'ify spapr_vio.c

Drop the old and empty SysBus init

Signed-off-by: xiaoqiang zhao <zxq_yx_007@163.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

hw/ppc: QOM'ify ppce500_spin.c

Drop the old SysBus init function and use instance_init

Signed-off-by: xiaoqiang zhao <zxq_yx_007@163.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

hw/ppc: QOM'ify e500.c

Drop the old SysBus init function and use instance_init

Signed-off-by: xiaoqiang zhao <zxq_yx_007@163.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

hw/gpio: QOM'ify mpc8xxx.c

* Drop the old SysBus init function and use instance_init
* Change mpc8xxx_gpio_reset to a DeviceClass::reset function

Signed-off-by: xiaoqiang zhao <zxq_yx_007@163.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

qtest: add ivshmem-test for ppc64

The test has been converted to use libqos, we can
now use it on ppc64. We also make the test fail on
all other architectures.
As libqos on ppc64 is not able to manage hotplug
and IRQ/MSI, we disable this part in the test on ppc64.

Signed-off-by: Laurent Vivier <lvivier@redhat.com>
[dwg: Make test conditional on CONFIG_EVENTFD]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

qtest: convert ivshmem-test to use libqos

This will allow to use it with ppc64.

Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

libqos: fix spapr qpci_map()

Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

qtest: add display-vga-test to ppc64

Only enable for ppc64 in the Makefile, but added
code in the file to check cirrus card only on architectures
supporting it (alpha, mips, i386, x86_64).

Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Tested-by: Greg Kurz <groug@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

qtest: add netfilter tests for ppc64

Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Tested-by: Greg Kurz <groug@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

ppc: Validate compatibility modes when setting

Current ppc_set_compat() will attempt to set any compatiblity mode
specified, regardless of whether it's available on the CPU.  The caller is
expected to make sure it is setting a possible mode, which is awkwward
because most of the information to make that decision is at the CPU level.

This begins to clean this up by introducing a ppc_check_compat() function
which will determine if a given compatiblity mode is supported on a CPU
(and also whether it lies within specified minimum and maximum compat
levels, which will be useful later).  It also contains an assertion that
the CPU has a "virtual hypervisor"[1], that is, that the guest isn't
permitted to execute hypervisor privilege code.  Without that, the guest
would own the PCR and so could override any mode set here.  Only machine
types which use a virtual hypervisor (i.e. 'pseries') should use
ppc_check_compat().

ppc_set_compat() is modified to validate the compatibility mode it is given
and fail if it's not available on this CPU.

[1] Or user-only mode, which also obviously doesn't allow access to the
hypervisor privileged PCR.  We don't use that now, but could in future.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>

ppc: Rewrite ppc_get_compat_smt_threads()

To continue consolidation of compatibility mode information, this rewrites
the ppc_get_compat_smt_threads() function using the table of compatiblity
modes in target-ppc/compat.c.

It's not a direct replacement, the new ppc_compat_max_threads() function
has simpler semantics - it just returns the number of threads the cpu
model has, taking into account any compatiblity mode it is in.

This no longer takes into account kvmppc_smt_threads() as the previous
version did.  That check wasn't useful because we check in
ppc_cpu_realizefn() that CPUs aren't instantiated with more threads
than kvm allows (or if we didn't things will already be broken and
this won't make it any worse).

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>

ppc: Rewrite ppc_set_compat()

This rewrites the ppc_set_compat() function so that instead of open coding
the various compatibility modes, it reads the relevant data from a table.
This is a first step in consolidating the information on compatibility
modes scattered across the code into a single place.

It also makes one change to the logic.  The old code masked the bits
to be set in the PCR (Processor Compatibility Register) by which bits
are valid on the host CPU.  This made no sense, since it was done
regardless of whether our guest CPU was the same as the host CPU or
not.  Furthermore, the actual PCR bits are only relevant for TCG[1] -
KVM instead uses the compatibility mode we tell it in
kvmppc_set_compat().  When using TCG host cpu information usually
isn't even present.

While we're at it, we put the new implementation in a new file to make the
enormous translate_init.c a little smaller.

[1] Actually it doesn't even do anything in TCG, but it will if / when we
    get to implementing compatibility mode logic at that level.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>

pseries: Add pseries-2.9 machine type

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>

prep: do not use global variable to access nvram

Signed-off-by: Hervé Poussineau <hpoussin@reactos.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

hw/ppc/spapr: Fix boot path of usb-host storage devices

When passing through an USB storage device to a pseries guest, it
is currently not possible to automatically boot from the device
if the "bootindex" property has been specified, too (e.g. when using
"-device nec-usb-xhci -device usb-host,hostbus=1,hostaddr=2,bootindex=0"
at the command line). The problem is that QEMU builds a device tree path
like "/pci@800000020000000/usb@0/usb-host@1" and passes it to SLOF
in the /chosen/qemu,boot-list property. SLOF, however, probes the
USB device, recognizes that it is a storage device and thus changes
its name to "storage", and additionally adds a child node for the
SCSI LUN, so the correct boot path in SLOF is something like
"/pci@800000020000000/usb@0/storage@1/disk@101000000000000" instead.
So when we detect an USB mass storage device with SCSI interface,
we've got to adjust the firmware boot-device path properly that
SLOF can automatically boot from the device.

Buglink: https://bugzilla.redhat.com/show_bug.cgi?id=1354177
Signed-off-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: implement stxvll instructions

stxvll: Store VSX Vector Left-justified with Length

Vector (8-bit elements) in BE/LE:
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+--+--+
|“T”|“h”|“i”|“s”|“ ”|“i”|“s”|“ ”|“a”|“ ”|“T”|“E”|“S”|“T”|00|00|
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+--+--+

Storing 14 bytes would result in following Little/Big-endian Storage:
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+--+--+
|“T”|“h”|“i”|“s”|“ ”|“i”|“s”|“ ”|“a”|“ ”|“T”|“E”|“S”|“T”|FF|FF|
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+--+--+

Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: implement stxvl instruction

stxvl: Store VSX Vector with Length

Vector (8-bit elements) in BE:
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+--+--+
|“T”|“h”|“i”|“s”|“ ”|“i”|“s”|“ ”|“a”|“ ”|“T”|“E”|“S”|“T”|00|00|
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+--+--+

Vector (8-bit elements) in LE:
+--+--+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
|00|00|“T”|“S”|“E”|“T”|“ ”|“a”|“ ”|“s”|“i”|“ ”|“s”|“i”|"h"|"T"|
+--+--+---+---+---+---+---+---+---+---+---+---+---+---+---+---+

Storing 14 bytes would result in following Little/Big-endian Storage:
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+--+--+
|“T”|“h”|“i”|“s”|“ ”|“i”|“s”|“ ”|“a”|“ ”|“T”|“E”|“S”|“T”|FF|FF|
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+--+--+

Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: implement lxvll instruction

lxvll: Load VSX Vector Left-justified with Length

Little/Big-endian Storage:
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+--+--+
|“T”|“h”|“i”|“s”|“ ”|“i”|“s”|“ ”|“a”|“ ”|“T”|“E”|“S”|“T”|FF|FF|
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+--+--+

Loading 14 bytes to vector (8-bit elements) in BE/LE:
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+--+--+
|“T”|“h”|“i”|“s”|“ ”|“i”|“s”|“ ”|“a”|“ ”|“T”|“E”|“S”|“T”|00|00|
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+--+--+

Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: implement lxvl instruction

lxvl: Load VSX Vector with Length

Little/Big-endian Storage:
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+--+--+
|“T”|“h”|“i”|“s”|“ ”|“i”|“s”|“ ”|“a”|“ ”|“T”|“E”|“S”|“T”|FF|FF|
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+--+--+

Loading 14 bytes results in:

Vector (8-bit elements) in BE:
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+--+--+
|“T”|“h”|“i”|“s”|“ ”|“i”|“s”|“ ”|“a”|“ ”|“T”|“E”|“S”|“T”|00|00|
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+--+--+

Vector (8-bit elements) in LE:
+--+--+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
|00|00|“T”|“S”|“E”|“T”|“ ”|“a”|“ ”|“s”|“i”|“ ”|“s”|“i”|"h"|"T"|
+--+--+---+---+---+---+---+---+---+---+---+---+---+---+---+---+

Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Add xxperm and xxpermr instructions

xxperm:  VSX Vector Permute
xxpermr: VSX Vector Permute Right-indexed

Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: implement xscpsgnqp instruction

xscpsgnqp: VSX Scalar Copy Sign Quad-Precision

Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: implement xsnegqp instruction

xsnegqp: VSX Scalar Negate Quad-Precision

Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Implement bcd_is_valid function

A function to check if all digits of a given BCD number is valid is
here presented because more instructions will need to reuse the
same code.

Signed-off-by: Jose Ricardo Ziviani <joserz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: implement xsabsqp/xsnabsqp instruction

xsabsqp:  VSX Scalar Absolute Quad-Precision
xsnabsqp: VSX Scalar Negative Absolute Quad-Precision

Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: implement stop instruction

Use the nap code.

Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: move ppc_vsr_t to common header

The structure and corresponding defines and functions need to be used
outside of fpu_helper.c as well.

Add u8, u16, u32 and Int128 to the structure.

Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

ppc/spapr: implement H_SIGNAL_SYS_RESET

The H_SIGNAL_SYS_RESET hcall allows a guest CPU to raise a system reset
exception on CPUs within the same guest -- all CPUs, all-but-self, or a
specific CPU (including self).

This has not made its way to a PAPR release yet, but we have an hcall
number assigned.

  H_SIGNAL_SYS_RESET = 0x380

  Syntax:
    hcall(uint64 H_SIGNAL_SYS_RESET, int64 target);

  Generate a system reset NMI on the threads indicated by target.

  Values for target:
    -1 = target all online threads including the caller
    -2 = target all online threads except for the caller
    All other negative values: reserved
    Positive values: The thread to be targeted, obtained from the value
    of the "ibm,ppc-interrupt-server#s" property of the CPU in the OF
    device tree.

  Semantics:
    - Invalid target: return H_Parameter.
    - Otherwise: Generate a system reset NMI on target thread(s),
      return H_Success.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

ppc: Rename cpu_version to compat_pvr

The 'cpu_version' field in PowerPCCPU is badly named.  It's named after the
'cpu-version' device tree property where it is advertised, but that meaning
may not be obvious in most places it appears.

Worse, it doesn't even really correspond to that device tree property.  The
property contains either the processor's PVR, or, if the CPU is running in
a compatibility mode, a special "logical PVR" representing which mode.

Rename the cpu_version field, and a number of related variables to
compat_pvr to make this clearer.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Thomas Huth <thuth@redhat.com>

ppc: Clean up and QOMify hypercall emulation

The pseries machine type is a bit unusual in that it runs a paravirtualized
guest.  The guest expects to interact with a hypervisor, and qemu
emulates the functions of that hypervisor directly, rather than executing
hypervisor code within the emulated system.

To implement this in TCG, we need to intercept hypercall instructions and
direct them to the machine's hypercall handlers, rather than attempting to
perform a privilege change within TCG.  This is controlled by a global
hook - cpu_ppc_hypercall.

This cleanup makes the handling a little cleaner and more extensible than
a single global variable.  Instead, each CPU to have hypercalls intercepted
has a pointer set to a QOM object implementing a new virtual hypervisor
interface.  A method in that interface is called by TCG when it sees a
hypercall instruction.  It's possible we may want to add other methods in
future.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>

pseries: Make cpu_update during CAS unconditional

spapr_h_cas_compose_response() includes a cpu_update parameter which
controls whether it includes updated information on the CPUs in the device
tree fragment returned from the ibm,client-architecture-support (CAS) call.

Providing the updated information is essential when CAS has negotiated
compatibility options which require different cpu information to be
presented to the guest.  However, it should be safe to provide in other
cases (it will just override the existing data in the device tree with
identical data).  This simplifies the code by removing the parameter and
always providing the cpu update information.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>

pseries: Always use core objects for CPU construction

Currently the pseries machine has two paths for constructing CPUs.  On
newer machine type versions, which support cpu hotplug, it constructs
cpu core objects, which in turn construct CPU threads.  For older machine
versions it individually constructs the CPU threads.

This division is going to make some future changes to the cpu construction
harder, so this patch unifies them.  Now cpu core objects are always
created.  This requires some updates to allow core objects to be created
without a full complement of threads (since older versions allowed a
number of cpus not a multiple of the threads-per-core).  Likewise it needs
some changes to the cpu core hot/cold plug path so as not to choke on the
old machine types without hotplug support.

For good measure, we move the cpu construction to its own subfunction,
spapr_init_cpus().

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>

target-ppc: add vextu[bhw][lr]x instructions

vextublx: Vector Extract Unsigned Byte Left
vextuhlx: Vector Extract Unsigned Halfword Left
vextuwlx: Vector Extract Unsigned Word Left
vextubrx: Vector Extract Unsigned Byte Right-Indexed VX-form
vextuhrx: Vector Extract Unsigned  Halfword Right-Indexed VX-form
vextuwrx: Vector Extract Unsigned Word Right-Indexed VX-form

Signed-off-by: Avinesh Kumar <avinesku@linux.vnet.ibm.com>
Signed-off-by: Hariharan T.S. <hari@linux.vnet.ibm.com>
[ implement using int128_rshift ]
Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Implement bcdsetsgn. instruction

bcdsetsgn.: Decimal set sign. This instruction copies the register
value to the result register but adjust the signal according to
the preferred sign value.

Signed-off-by: Jose Ricardo Ziviani <joserz@linux.vnet.ibm.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Implement bcdcpsgn. instruction

bcdcpsgn.: Decimal copy sign. Given two registers vra and vrb, it
copies the vra value with vrb sign to the result register vrt.

Signed-off-by: Jose Ricardo Ziviani <joserz@linux.vnet.ibm.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Implement bcdctsq. instruction

bcdctsq.: Decimal convert to signed quadword. It is possible to
convert packed decimal values to signed quadwords.

Signed-off-by: Jose Ricardo Ziviani <joserz@linux.vnet.ibm.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Implement bcdcfsq. instruction

bcdcfsq.: Decimal convert from signed quadword. It is not possible
to convert values less than -10^31-1 or greater than 10^31-1 to be
represented in packed decimal format.

Signed-off-by: Jose Ricardo Ziviani <joserz@linux.vnet.ibm.com>
[dwg: Corrected constant which should be 10^16-1 but was 10^17-1]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: implement lxv/lxvx and stxv/stxvx

lxv:  Load VSX Vector
lxvx: Load VSX Vector Indexed

    Little/Big-endian Storage
    +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
    |F0|F1|F2|F3|F4|F5|F6|F7|E0|E1|E2|E3|E4|E5|E6|E7|
    +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+

    Vector load results:
    BE:
    +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
    |F0|F1|F2|F3|F4|F5|F6|F7|E0|E1|E2|E3|E4|E5|E6|E7|
    +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+

    LE:
    +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
    |E7|E6|E5|E4|E3|E2|E1|E0|F7|F6|F5|F4|F3|F2|F1|F0|
    +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+

stxv: Store VSX Vector
stxvx: Store VSX Vector Indexed

    Vector (8-bit elements) in BE:
    +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
    |F0|F1|F2|F3|F4|F5|F6|F7|E0|E1|E2|E3|E4|E5|E6|E7|
    +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+

    Vector (8-bit elements) in LE:
    +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
    |E7|E6|E5|E4|E3|E2|E1|E0|F7|F6|F5|F4|F3|F2|F1|F0|
    +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+

    Store results in following:
    +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
    |F0|F1|F2|F3|F4|F5|F6|F7|E0|E1|E2|E3|E4|E5|E6|E7|
    +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+

Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: implement stxsd and stxssp

stxsd:  Store VSX Scalar Dword
stxssp: Store VSX Scalar SP

Moreover, DQ-Form/DS-FORM instructions shares the same primary
opcode(0x3D). For DQ-FORM bits 29:31 are used, for DS-FORM bits 30:31
are used. Common routine to decode primary opcode(0x3D) -
ds-form/dq-form instructions is required.

Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: implement lxsd and lxssp instructions

lxsd: Load VSX Scalar Dword
lxssp: Load VSX Scalar Single

Moreover, DS-Form instructions shares the same primary opcode, bits
30:31 are used to decode the instruction. Use a common routine to decode
primary opcode(0x39) - ds-form instructions and branch-out depending on
bits 30:31.

Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Add xscmpoqp and xscmpuqp instructions

xscmpoqp - VSX Scalar Compare Ordered Quad-Precision
xscmpuqp - VSX Scalar Compare Unordered Quad-Precision

Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Add xscmpexp[dp,qp] instructions

xscmpexpdp: VSX Scalar Compare Exponents Double-Precision
xscmpexpqp: VSX Scalar Compare Exponents Quad-Precision

Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Fix xscmpodp and xscmpudp instructions

- xscmpodp & xscmpudp are missing flags reset.
- In xscmpodp, VXCC should be set only if VE is 0 for signalling NaN case
  and VXCC should be set by explicitly checking for quiet NaN case.
- Comparison is being done only if the operands are not NaNs. However as
  per ISA, it should be done even when operands are NaNs.

Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: rename CRF_* defines as CRF_*_BIT

Add _BIT to CRF_[GT,LT,EQ_SO] and introduce CRF_[GT,LT,EQ,SO] for usage
without shifts in the code. This would simplify the code.

Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

target-ppc: Consolidate instruction decode helpers

Move instruction decode helpers to target-ppc/internal.h so that some
of these can be used from outside of translate.c. This movement also
helps to get rid of some duplicate helpers from target-ppc/fpu_helper.c.

Suggested-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Signed-off-by: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

disas/ppc: Fix indefinite articles in comments

Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

Merge remote-tracking branch 'remotes/bonzini/tags/for-upstream' into staging

* SCSI max_transfer support for scsi-generic (Eric)
* x86 SMI broadcast (Laszlo)
* Character device QOMification (Marc-André)
* Record/replay improvements (Pavel)
* iscsi fixes (Peter L.)
* "info mtree -f" command (Peter Xu)
* TSC clock rate reporting (Phil)
* DEVICE_CATEGORY_CPU (Thomas)
* Memory sign-extension fix (Ladi)

# gpg: Signature made Fri 27 Jan 2017 17:08:51 GMT
# gpg:                using RSA key 0xBFFBD25F78C7AE83
# gpg: Good signature from "Paolo Bonzini <bonzini@gnu.org>"
# gpg:                 aka "Paolo Bonzini <pbonzini@redhat.com>"
# Primary key fingerprint: 46F5 9FBD 57D6 12E7 BFD4  E2F7 7E15 100C CD36 69B1
#      Subkey fingerprint: F133 3857 4B66 2389 866C  7682 BFFB D25F 78C7 AE83

* remotes/bonzini/tags/for-upstream: (41 commits)
  memory: don't sign-extend 32-bit writes
  chardev: qom-ify
  vc: use a common prefix for chr callbacks
  baum: use a common prefix for chr callbacks
  gtk: overwrite the console.c char driver
  char: use error_report()
  spice-char: improve error reporting
  char: rename TCPChardev and NetChardev
  char: rename CharDriverState Chardev
  bt: use qemu_chr_alloc()
  char: allocate CharDriverState as a single object
  char: use a feature bit for replay
  char: introduce generic qemu_chr_get_kind()
  char: fold single-user functions in caller
  char: move callbacks in CharDriver
  char: use a static array for backends
  char: use a const CharDriver
  doc: fix spelling
  char: add qemu_chr_fe_add_watch() Returns description
  qemu-options: stdio is available on win32
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

memory: don't sign-extend 32-bit writes

ldl_p has a signed return type so assigning it to uint64_t implicitly
sign-extends the value. This results in devices with min_access_size = 8
seeing unexpected values passed to their write handlers.

Example: guest performs a 32-bit write of 0x80000000 to an mmio region
and the handler receives 0xFFFFFFFF80000000 in its value argument.

Signed-off-by: Ladi Prosek <lprosek@redhat.com>
Message-Id: <1485440557-10384-1-git-send-email-lprosek@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

chardev: qom-ify

Turn Chardev into Object.

qemu_chr_alloc() is replaced by the qemu_chardev_new() constructor. It
will call qemu_char_open() to open/intialize the chardev with the
ChardevCommon *backend settings.

The CharDriver::create() callback is turned into a ChardevClass::open()
which is called from the newly introduced qemu_chardev_open().

"chardev-gdb" and "chardev-hci" are internal chardev and aren't
creatable directly with -chardev. Use a new internal flag to disable
them. We may want to use TYPE_USER_CREATABLE interface instead, or
perhaps allow -chardev usage.

Although in general we keep typename and macros private, unless the type
is being used by some other file, in this patch, all types and common
helper macros for qemu-char.c are in char.h. This is to help transition
now (some types must be declared early, while some aren't shared) and
when splitting in several units. This is to be improved later.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

vc: use a common prefix for chr callbacks

vc_chr_write() is more appropriate than _puts() since no newline is
appended, even though it's not used only as a callback.

Keep "qemu_chr_parse" prefix, most chardev parse functions use this
prefix atm.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

baum: use a common prefix for chr callbacks

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

gtk: overwrite the console.c char driver

Instead of registering a vc handler to allocate the Gtk VC Chardev,
overwrite the console.c char driver.

A later patch, when switching to QOM, will register a default console vc
QOM class if none has been registered before.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

char: use error_report()

Prefer error_report() over fprintf(stderr..)

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

spice-char: improve error reporting

Set errp to report errors up to the right monitor.

Use error_append_hint() to give hints about parameters on !qmp monitors,
instead of a direct fprintf() call.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

char: rename TCPChardev and NetChardev

Rename the types to follow the name of the chardev kind.
- socket: TCPChardev -> SocketChardev
- udp: NetChardev -> UdpChardev

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

char: rename CharDriverState Chardev

Pick a uniform chardev type name.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

bt: use qemu_chr_alloc()

Use common allocator for CharDriverState.

Rename the now untouched parent field.

The casts added are temporary, they are replaced with QOM type-safe
macros in a later patch in this series.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

char: allocate CharDriverState as a single object

Use a single allocation for CharDriverState, this avoids extra
allocations & pointers, and is a step towards more object-oriented
CharDriver.

Gtk console is a bit peculiar, gd_vc_chr_set_echo() used to have a
temporary VirtualConsole to save the echo bit. Instead now, we consider
whether vcd->console is set or not, and restore the echo bit saved in
VCDriverState when calling gd_vc_vte_init().

The casts added are temporary, they are replaced with QOM type-safe
macros in a later patch in this series.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

char: use a feature bit for replay

Use a feature flag rather than a structure field for "replay".

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

char: introduce generic qemu_chr_get_kind()

This allows to remove the "is_mux" field from CharDriverState.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

char: fold single-user functions in caller

This shortens the code a bit.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

char: move callbacks in CharDriver

This makes the code more declarative, and avoids duplicating the
information on all instances.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

char: use a static array for backends

Number and kinds of backends is known at compile-time, use a fixed-sized
static array to simplify iterations & lookups.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

char: use a const CharDriver

No need to allocate & copy fields, let's use static const struct instead.

Add an alias field to the CharDriver structure to cover the cases where
we previously registered a driver twice under two names.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

doc: fix spelling

I am pretty sure that's the word Fabrice Bellard intended to write.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

char: add qemu_chr_fe_add_watch() Returns description

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

qemu-options: stdio is available on win32

Available since commit db418a0a7ef5887ea0f3d167584e6f500bb0c4c5
(October 2011, qemu 1.0)

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

tests: fix linking test-char on win32

test.char.exe fails to link:
qemu-char.o: In function `win_chr_free':
/home/elmarco/src/qemu/qemu-char.c:2149: undefined reference to `qemu_del_polling_cb'
/home/elmarco/src/qemu/qemu-char.c:2151: undefined reference to `qemu_del_polling_cb'
qemu-char.o: In function `win_stdio_thread':
/home/elmarco/src/qemu/qemu-char.c:2568: undefined reference to `qemu_del_wait_object'
qemu-char.o: In function `qemu_chr_open_stdio':
/home/elmarco/src/qemu/qemu-char.c:2661: undefined reference to `qemu_add_wait_object'
/home/elmarco/src/qemu/qemu-char.c:2646: undefined reference to
`qemu_add_wait_object'
...

It needs main-loop.o symbols, among others. Linking with
$(test-block-obj-y) brings what's necessary. We could try to eventually
strip to the minimum if needed.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

block/iscsi: statically link qemu_iscsi_opts

commit f57b4b5f moved qemu_iscsi_opts into vl.c. This
made them invisible for qemu-img, qemu-nbd etc.

Fixes: f57b4b5fb127b60e1aade2684a8b16bc4f630b29
Cc: qemu-stable@nongnu.org
Signed-off-by: Peter Lieven <pl@kamp.de>
Message-Id: <1485262161-18543-1-git-send-email-pl@kamp.de>
[Drop useless #ifdef. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

pc: Enable vmware-cpuid-freq CPU option for 2.9+ machine types

Signed-off-by: Phil Dennis-Jordan <phil@philjordan.eu>
Message-Id: <1484921496-11257-4-git-send-email-phil@philjordan.eu>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

x86-KVM: Supply TSC and APIC clock rates to guest like VMWare

This fixes timekeeping of x86-64 Darwin/OS X/macOS guests when using KVM.

Darwin/OS X/macOS for x86-64 uses the TSC for timekeeping; it normally calibrates this by querying various clock frequency scaling MSRs. Details depend on the exact CPU model detected. The local APIC timer frequency is extracted from (EFI) firmware.

This is problematic in the presence of virtualisation, as the MSRs in question are typically not handled by the hypervisor. VMWare (Fusion) advertises TSC and APIC frequency via a custom 0x40000010 CPUID leaf, in the eax and ebx registers respectively. This is documented at https://lwn.net/Articles/301888/ among other places.

Darwin/OS X/macOS looks for the generic 0x40000000 hypervisor leaf, and if this indicates via eax that leaf 0x40000010 might be available, that is in turn queried for the two frequencies.

This adds a CPU option "vmware-cpuid-freq" to enable the same behaviour when running Qemu with KVM acceleration, if the KVM TSC frequency can be determined, and it is stable. (invtsc or user-specified) The virtualised APIC bus cycle is hardcoded to 1GHz in KVM, so ebx of the CPUID leaf is also hardcoded to this value.

Signed-off-by: Phil Dennis-Jordan <phil@philjordan.eu>
Message-Id: <1484921496-11257-2-git-send-email-phil@philjordan.eu>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

block: get max_transfer limit for char (scsi-generic) devices

We can get the maximum number of bytes for a single I/O transfer
from the BLKSECTGET ioctl, but we only perform this for block
devices.  scsi-generic devices are represented as character devices,
and so do not issue this today.  Update this, so that virtio-scsi
devices using the scsi-generic interface can return the same data.

Signed-off-by: Eric Farman <farman@linux.vnet.ibm.com>
Message-Id: <20170120162527.66075-4-farman@linux.vnet.ibm.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

block: Fix target variable of BLKSECTGET ioctl

Commit 6f6071745bd0 ("raw-posix: Fetch max sectors for host block device")
introduced a routine to call the kernel BLKSECTGET ioctl, which stores the
result back to user space.  However, the size of the data returned depends
on the routine handling the ioctl.  The (compat_)blkdev_ioctl returns a
short, while sg_ioctl returns an int.  Thus, on big-endian systems, we can
find ourselves accidentally shifting the result to a much larger value.
(On s390x, a short is 16 bits while an int is 32 bits.)

Also, the two ioctl handlers return values in different scales (block
returns sectors, while sg returns bytes), so some tweaking of the outputs
is required such that hdev_get_max_transfer_length returns a value in a
consistent set of units.

Signed-off-by: Eric Farman <farman@linux.vnet.ibm.com>
Message-Id: <20170120162527.66075-3-farman@linux.vnet.ibm.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

hw/scsi: Fix debug message of cdb structure in scsi-generic

When running with debug enabled, the scsi-generic cdb that is
dumped skips byte 0 of the command, which is the opcode.  This
makes identifying which command is being issued/completed a
little difficult.  Example:

  0x00 0x00 0x01 0x00 0x00
  scsi-generic: scsi_read_data 0x0
  scsi-generic: Data ready tag=0x0 len=164
  scsi-generic: scsi_read_data 0x0
  scsi-generic: Command complete 0x0x10a42c60 tag=0x0 status=0

Improve this by adding a message prior to the loop, similar to
what exists for scsi-disk.  Clean up a few other messages to be
more explicit of what is being represented.  Example:

  scsi-generic: Command: data=0x12 0x00 0x00 0x01 0x00 0x00
  scsi-generic: scsi_read_data tag=0x0
  scsi-generic: Data ready tag=0x0 len=164
  scsi-generic: scsi_read_data tag=0x0
  scsi-generic: Command complete 0x0x10a452d0 tag=0x0 status=0

Signed-off-by: Eric Farman <farman@linux.vnet.ibm.com>
Message-Id: <20170120162527.66075-2-farman@linux.vnet.ibm.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Introduce DEVICE_CATEGORY_CPU for CPU devices

Now that CPUs show up in the help text of "-device ?",
we should group them into an appropriate category.

Signed-off-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <1484917276-7107-1-git-send-email-thuth@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

block/iscsi: avoid data corruption with cache=writeback

nb_cls_shrunk in iscsi_allocmap_update can become -1 if the
request starts and ends within the same cluster. This results
in passing -1 to bitmap_set and bitmap_clear and they don't
handle negative values properly. In the end this leads to data
corruption.

Fixes: e1123a3b40a1a9a625a29c8ed4debb7e206ea690
Cc: qemu-stable@nongnu.org
Signed-off-by: Peter Lieven <pl@kamp.de>
Message-Id: <1484579832-18589-1-git-send-email-pl@kamp.de>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

hw/isa/lpc_ich9: negotiate SMI broadcast on pc-q35-2.9+ machine types

Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Eduardo Habkost <ehabkost@redhat.com>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Cc: Igor Mammedov <imammedo@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Message-Id: <20170126014416.11211-4-lersek@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

hw/isa/lpc_ich9: add broadcast SMI feature

The generic edk2 SMM infrastructure prefers
EFI_SMM_CONTROL2_PROTOCOL.Trigger() to inject an SMI on each processor. If
Trigger() only brings the current processor into SMM, then edk2 handles it
in the following ways:

(1) If Trigger() is executed by the BSP (which is guaranteed before
    ExitBootServices(), but is not necessarily true at runtime), then:

    (a) If edk2 has been configured for "traditional" SMM synchronization,
        then the BSP sends directed SMIs to the APs with APIC delivery,
        bringing them into SMM individually. Then the BSP runs the SMI
        handler / dispatcher.

    (b) If edk2 has been configured for "relaxed" SMM synchronization,
        then the APs that are not already in SMM are not brought in, and
        the BSP runs the SMI handler / dispatcher.

(2) If Trigger() is executed by an AP (which is possible after
    ExitBootServices(), and can be forced e.g. by "taskset -c 1
    efibootmgr"), then the AP in question brings in the BSP with a
    directed SMI, and the BSP runs the SMI handler / dispatcher.

The smaller problem with (1a) and (2) is that the BSP and AP
synchronization is slow. For example, the "taskset -c 1 efibootmgr"
command from (2) can take more than 3 seconds to complete, because
efibootmgr accesses non-volatile UEFI variables intensively.

The larger problem is that QEMU's current behavior diverges from the
behavior usually seen on physical hardware, and that keeps exposing
obscure corner cases, race conditions and other instabilities in edk2,
which generally expects / prefers a software SMI to affect all CPUs at
once.

Therefore introduce the "broadcast SMI" feature that causes QEMU to inject
the SMI on all VCPUs.

While the original posting of this patch
<http://lists.nongnu.org/archive/html/qemu-devel/2015-10/msg05658.html>
only intended to speed up (2), based on our recent "stress testing" of SMM
this patch actually provides functional improvements.

Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Cc: Igor Mammedov <imammedo@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Message-Id: <20170126014416.11211-3-lersek@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

hw/isa/lpc_ich9: add SMI feature negotiation via fw_cfg

Introduce the following fw_cfg files:

- "etc/smi/supported-features": a little endian uint64_t feature bitmap,
  presenting the features known by the host to the guest. Read-only for
  the guest.

  The content of this file will be determined via bit-granularity ICH9-LPC
  device properties, to be introduced later. For now, the bitmask is left
  zeroed. The bits will be set from machine type compat properties and on
  the QEMU command line, hence this file is not migrated.

- "etc/smi/requested-features": a little endian uint64_t feature bitmap,
  representing the features the guest would like to request. Read-write
  for the guest.

  The guest can freely (re)write this file, it has no direct consequence.
  Initial value is zero. A nonzero value causes the SMI-related fw_cfg
  files and fields that are under guest influence to be migrated.

- "etc/smi/features-ok": contains a uint8_t value, and it is read-only for
  the guest. When the guest selects the associated fw_cfg key, the guest
  features are validated against the host features. In case of error, the
  negotiation doesn't proceed, and the "features-ok" file remains zero. In
  case of success, the "features-ok" file becomes (uint8_t)1, and the
  negotiated features are locked down internally (to which no further
  changes are possible until reset).

  The initial value is zero.  A nonzero value causes the SMI-related
  fw_cfg files and fields that are under guest influence to be migrated.

The C-language fields backing the "supported-features" and
"requested-features" files are uint8_t arrays. This is because they carry
guest-side representation (our choice is little endian), while
VMSTATE_UINT64() assumes / implies host-side endianness for any uint64_t
fields. If we migrate a guest between hosts with different endiannesses
(which is possible with TCG), then the host-side value is preserved, and
the host-side representation is translated. This would be visible to the
guest through fw_cfg, unless we used plain byte arrays. So we do.

Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Cc: Igor Mammedov <imammedo@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Message-Id: <20170126014416.11211-2-lersek@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

memory: hmp: add "-f" for "info mtree"

Adding one more option "-f" for "info mtree" to dump the flat views of
all the address spaces.

This will be useful to debug the memory rendering logic, also it'll be
much easier with it to know what memory region is handling what address
range.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <1484556005-29701-3-git-send-email-peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

memory: tune mtree_print_mr() to dump mr type

We were dumping RW bits for each memory region, that might be confusing.
It'll make more sense to dump the memory region type directly rather
than the RW bits since that's how the bits are derived.

Meanwhile, with some slight cleanup in the function.

Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <1484556005-29701-2-git-send-email-peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

apic: save apic_delivered flag

This patch implements saving/restoring of static apic_delivered variable.

v8: saving static variable only for one of the APICs

Signed-off-by: Pavel Dovgalyuk <pavel.dovgaluk@ispras.ru>
Message-Id: <20170126123429.5412.94368.stgit@PASHA-ISP>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

replay: exception replay fix

This patch fixes replaying the exception when TB cache is full.
It breaks cpu loop execution through setting exception_index
to process such queued work as TB flush.

v8: moved setting of exeption_index to tb_gen_code

Signed-off-by: Pavel Dovgalyuk <pavel.dovgaluk@ispras.ru>
Message-Id: <20170126123418.5412.33815.stgit@PASHA-ISP>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

replay: save/load initial state

This patch implements initial vmstate creation or loading at the start
of record/replay. It is needed for rewinding the execution in the replay mode.

v4 changes:
 - snapshots are not created by default anymore

v3 changes:
 - added rrsnapshot option

Signed-off-by: Pavel Dovgalyuk <pavel.dovgaluk@ispras.ru>
Message-Id: <20170124071746.4572.61449.stgit@PASHA-ISP>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

savevm: add public save_vmstate function

This patch introduces save_vmstate function to allow saving and loading
vmstates from the replay module.

Signed-off-by: Pavel Dovgalyuk <pavel.dovgaluk@ispras.ru>
Message-Id: <20170124071741.4572.13714.stgit@PASHA-ISP>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>