qemu.git
5 years agohw: Add support for LSI SAS1068 (mptsas) device
Paolo Bonzini [Tue, 27 Oct 2015 17:26:06 +0000 (18:26 +0100)] 
hw: Add support for LSI SAS1068 (mptsas) device

This adds the SAS1068 device, a SAS disk controller used in VMware that
is oldish but widely supported and has decent performance.  Unlike
megasas, it presents itself as a SAS controller and not as a RAID
controller.  The device corresponds to the mptsas kernel driver in
Linux.

A few small things in the device setup are based on Don Slutz's old
patch, but the device emulation was written from scratch based on Don's
SeaBIOS patch and on the FreeBSD and Linux drivers.  It is 2400 lines
shorter than Don's patch (and roughly the same size as MegaSAS---also
because it doesn't support the similar SPI controller), implements SCSI
task management functions (with asynchronous cancellation), supports
big-endian hosts, has complete support for migration and follows the
QEMU coding standards much more closely.

To write the driver, I first split Don's patch in two parts, with
the configuration bits in one file and the rest in a separate file.
I first left mptconfig.c in place and rewrote the rest, then deleted
mptconfig.c as well.  The configuration pages are still based mostly on
VirtualBox's, though not exactly the same.  However, the implementation
is completely different.  The contents of the pages themselves should
not be copyrightable.

Signed-off-by: Don Slutz <Don@CloudSwitch.com>
Message-Id: <1347382813-5662-1-git-send-email-Don@CloudSwitch.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoscsi-generic: grab device and port SAS addresses from backend
Paolo Bonzini [Tue, 12 Jan 2016 11:29:57 +0000 (12:29 +0100)] 
scsi-generic: grab device and port SAS addresses from backend

This lets a SAS adapter expose them through its own configuration
mechanism.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoscsi: push WWN fields up to SCSIDevice
Paolo Bonzini [Tue, 12 Jan 2016 10:47:00 +0000 (11:47 +0100)] 
scsi: push WWN fields up to SCSIDevice

SAS adapters need to access them in order to publish the SAS addresses
of the end devices connected to them.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoinclude/qemu/atomic.h: default to __atomic functions
Alex Bennée [Thu, 28 Jan 2016 10:15:17 +0000 (10:15 +0000)] 
include/qemu/atomic.h: default to __atomic functions

The __atomic primitives have been available since GCC 4.7 and provide
a richer interface for describing memory ordering requirements. As a
bonus by using the primitives instead of hand-rolled functions we can
use tools such as the ThreadSanitizer which need the use of well
defined APIs for its analysis.

If we have __ATOMIC defines we exclusively use the __atomic primitives
for all our atomic access. Otherwise we fall back to the mixture of
__sync and hand-rolled barrier cases.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <1453976119-24372-4-git-send-email-alex.bennee@linaro.org>
[Use __ATOMIC_SEQ_CST for atomic_mb_read/atomic_mb_set on !POWER. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoconfigure: sanity check the glib library that pkg-config finds
Daniel P. Berrange [Wed, 27 Jan 2016 09:00:45 +0000 (09:00 +0000)] 
configure: sanity check the glib library that pkg-config finds

Developers on 64-bit machines will often try to perform a
32-bit build of QEMU by running

  ./configure --extra-cflags="-m32"

Unfortunately if PKG_CONFIG_LIBDIR is not set to point to
the location of the 32-bit pkg-config files, then configure
will silently pick up the 64-bit pkg-config files and still
succeed.

This causes a problem for glib because it means QEMU will
be pulling in /usr/lib64/glib-2.0/include/glibconfig.h
instead of /usr/lib/glib-2.0/include/glibconfig.h

This causes problems because the 'gsize' type (defined as
'unsigned long') will no longer be fully compatible with
the 'size_t' type (defined as 'unsigned int'). Although
both are the same size, the compiler refuses to allow
casts from 'unsigned long *' to 'unsigned int *' as they
are different pointer types. This results in non-obvious
compiler errors when building QEMU eg

qga/commands-posix.c: In function â€˜qmp_guest_set_user_password’:
qga/commands-posix.c:1912:55: error: passing argument 2 of â€˜g_base64_decode’ from incompatible pointer type [-Werror=incompatible-pointer-types]
     rawpasswddata = (char *)g_base64_decode(password, &rawpasswdlen);
                                                            ^
In file included from /usr/include/glib-2.0/glib.h:35:0,
                 from qga/commands-posix.c:14:
/usr/include/glib-2.0/glib/gbase64.h:52:9: note: expected â€˜gsize * {aka long unsigned int *}’ but argument is of type â€˜size_t * {aka unsigned int *}’
 guchar *g_base64_decode         (const gchar  *text,
         ^
cc1: all warnings being treated as errors

To detect this problem, add a check to configure that
verifies that GLIB_SIZEOF_SIZE_T matches sizeof(size_t).
If this fails print a warning suggesting that the dev
probably needs to set PKG_CONFIG_LIBDIR.

On Fedora x86_64 it passes with any of:

 # ./configure
 # PKG_CONFIG_LIBDIR=/usr/lib/pkgconfig ./configure --extra-cflags="-m32"
 # PKG_CONFIG_LIBDIR=/usr/lib64/pkgconfig ./configure --extra-cflags="-m64"

And fails with a mis-match

 # PKG_CONFIG_LIBDIR=/usr/lib64/pkgconfig ./configure --extra-cflags="-m32"
 # PKG_CONFIG_LIBDIR=/usr/lib/pkgconfig ./configure --extra-cflags="-m64"

ERROR: sizeof(size_t) doesn't match GLIB_SIZEOF_SIZE_T.
       You probably need to set PKG_CONFIG_LIBDIR
       to point to the right pkg-config files for your
       build target

Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Message-Id: <1453885245-15562-1-git-send-email-berrange@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoqemu-char: Keep pty slave file descriptor open until the master is closed
Paolo Bonzini [Tue, 19 Jan 2016 13:13:29 +0000 (14:13 +0100)] 
qemu-char: Keep pty slave file descriptor open until the master is closed

If a process opens the slave pts device, writes data to it, then
immediately closes it, the data doesn't reliably get delivered to the
emulated serial port. This seems to be because a read of the master
pty device returns EIO on Linux if no process has the pts device open,
even when data is waiting "in the pipe".

A fix seems to be for QEMU to keep the pts file descriptor open until
the pty is closed, as per the below patch.

Signed-off-by: Ashley Jonathan <jonathan.ashley@altran.com>
Message-Id: <AC19797808C8D548ABDE0CA4A97AA30A30DEB409@XMB-DCFR-37.europe.corp.altran.com>
Reviewed-by: Michael Tokarev <mjt@tls.msk.ru>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agomemory: RCU ram_list.dirty_memory[] for safe RAM hotplug
Stefan Hajnoczi [Mon, 25 Jan 2016 13:33:20 +0000 (13:33 +0000)] 
memory: RCU ram_list.dirty_memory[] for safe RAM hotplug

Although accesses to ram_list.dirty_memory[] use atomics so multiple
threads can safely dirty the bitmap, the data structure is not fully
thread-safe yet.

This patch handles the RAM hotplug case where ram_list.dirty_memory[] is
grown.  ram_list.dirty_memory[] is change from a regular bitmap to an
RCU array of pointers to fixed-size bitmap blocks.  Threads can continue
accessing bitmap blocks while the array is being extended.  See the
comments in the code for an in-depth explanation of struct
DirtyMemoryBlocks.

I have tested that live migration with virtio-blk dataplane works.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <1453728801-5398-2-git-send-email-stefanha@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agomemory: add early bail out from cpu_physical_memory_set_dirty_range
Paolo Bonzini [Mon, 25 Jan 2016 14:13:47 +0000 (15:13 +0100)] 
memory: add early bail out from cpu_physical_memory_set_dirty_range

This condition is true in the common case, so we can cut out the body of
the function.  In addition, this makes it easier for the compiler to do
at least partial inlining, even if it decides that fully inlining the
function is unreasonable.

Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoMerge remote-tracking branch 'remotes/armbru/tags/pull-qapi-2016-02-09' into staging
Peter Maydell [Tue, 9 Feb 2016 11:42:43 +0000 (11:42 +0000)] 
Merge remote-tracking branch 'remotes/armbru/tags/pull-qapi-2016-02-09' into staging

QAPI patches for 2016-02-09

# gpg: Signature made Tue 09 Feb 2016 10:55:51 GMT using RSA key ID EB918653
# gpg: Good signature from "Markus Armbruster <armbru@redhat.com>"
# gpg:                 aka "Markus Armbruster <armbru@pond.sub.org>"

* remotes/armbru/tags/pull-qapi-2016-02-09: (31 commits)
  qapi: Add missing JSON files in build dependencies
  qapi: Fix compilation failure on MIPS and SPARC
  qmp: Don't abuse stack to track qmp-output root
  qmp: Fix reference-counting of qnull on empty output visit
  qapi: Drop unused error argument for list and implicit struct
  qapi: Tighten qmp_input_end_list()
  qapi: Drop unused 'kind' for struct/enum visit
  qapi: Swap 'name' in visit_* callbacks to match public API
  qom: Swap 'name' next to visitor in ObjectPropertyAccessor
  qapi: Swap visit_* arguments for consistent 'name' placement
  qom: Use typedef for Visitor
  qapi: Don't cast Enum* to int*
  qapi: Consolidate visitor small integer callbacks
  qapi: Make all visitors supply uint64 callbacks
  qapi: Prefer type_int64 over type_int in visitors
  qapi-visit: Kill unused visit_end_union()
  qapi: Track all failures between visit_start/stop
  qapi: Improve generated event use of qapi visitor
  balloon: Improve use of qapi visitor
  vl: Ensure qapi visitor properly ends struct visit
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
5 years agoMerge remote-tracking branch 'remotes/rth/tags/pull-tcg-20160209' into staging
Peter Maydell [Tue, 9 Feb 2016 09:22:23 +0000 (09:22 +0000)] 
Merge remote-tracking branch 'remotes/rth/tags/pull-tcg-20160209' into staging

Queued TCG patches

# gpg: Signature made Mon 08 Feb 2016 23:57:30 GMT using RSA key ID 4DD0279B
# gpg: Good signature from "Richard Henderson <rth7680@gmail.com>"
# gpg:                 aka "Richard Henderson <rth@redhat.com>"
# gpg:                 aka "Richard Henderson <rth@twiddle.net>"

* remotes/rth/tags/pull-tcg-20160209:
  tcg: Introduce temp_load
  tcg: Change temp_save argument to TCGTemp
  tcg: Change temp_sync argument to TCGTemp
  tcg: Change temp_dead argument to TCGTemp
  tcg: Change reg_to_temp to TCGTemp pointer
  tcg: Remove tcg_get_arg_str_i32/64
  tcg: More use of TCGReg where appropriate
  tcg: Work around clang bug wrt enum ranges
  tcg: Tidy temporary allocation
  tcg: Change ts->mem_reg to ts->mem_base
  tcg: Change tcg_global_mem_new_* to take a TCGv_ptr
  tcg: Remove lingering references to gen_opc_buf
  tcg: Respect highwater in tcg_out_tb_finalize

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
5 years agotcg: Introduce temp_load
Richard Henderson [Thu, 19 Sep 2013 15:02:05 +0000 (08:02 -0700)] 
tcg: Introduce temp_load

Unify all of the places that realize a temporary into a register.

Reviewed-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
5 years agotcg: Change temp_save argument to TCGTemp
Richard Henderson [Wed, 18 Sep 2013 22:35:32 +0000 (15:35 -0700)] 
tcg: Change temp_save argument to TCGTemp

Reviewed-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
5 years agotcg: Change temp_sync argument to TCGTemp
Richard Henderson [Wed, 18 Sep 2013 22:33:00 +0000 (15:33 -0700)] 
tcg: Change temp_sync argument to TCGTemp

Reviewed-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
5 years agotcg: Change temp_dead argument to TCGTemp
Richard Henderson [Wed, 18 Sep 2013 22:29:18 +0000 (15:29 -0700)] 
tcg: Change temp_dead argument to TCGTemp

Reviewed-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
5 years agotcg: Change reg_to_temp to TCGTemp pointer
Richard Henderson [Wed, 18 Sep 2013 22:21:56 +0000 (15:21 -0700)] 
tcg: Change reg_to_temp to TCGTemp pointer

Reviewed-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
5 years agotcg: Remove tcg_get_arg_str_i32/64
Richard Henderson [Wed, 18 Sep 2013 22:08:52 +0000 (15:08 -0700)] 
tcg: Remove tcg_get_arg_str_i32/64

Reviewed-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
5 years agotcg: More use of TCGReg where appropriate
Richard Henderson [Wed, 18 Sep 2013 21:54:45 +0000 (14:54 -0700)] 
tcg: More use of TCGReg where appropriate

Reviewed-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
5 years agotcg: Work around clang bug wrt enum ranges
Richard Henderson [Mon, 8 Feb 2016 23:43:42 +0000 (10:43 +1100)] 
tcg: Work around clang bug wrt enum ranges

A subsequent patch patch will change the type of REG from int
to enum TCGReg, which provokes the following bug in clang:

  https://llvm.org/bugs/show_bug.cgi?id=16154

Signed-off-by: Richard Henderson <rth@twiddle.net>
5 years agotcg: Tidy temporary allocation
Richard Henderson [Thu, 19 Sep 2013 15:46:21 +0000 (08:46 -0700)] 
tcg: Tidy temporary allocation

In particular, make sure the memory is memset before use.
Continues the increased use of TCGTemp pointers instead of
integer indices where appropriate.

Reviewed-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
5 years agotcg: Change ts->mem_reg to ts->mem_base
Richard Henderson [Wed, 18 Sep 2013 21:12:53 +0000 (14:12 -0700)] 
tcg: Change ts->mem_reg to ts->mem_base

Chain the temporaries together via pointers intstead of indices.
The mem_reg value is now mem_base->reg.  This will be important later.

This does require that the frame pointer have a global temporary
allocated for it.  This is simple bar the existing reserved_regs check.

Reviewed-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
5 years agotcg: Change tcg_global_mem_new_* to take a TCGv_ptr
Richard Henderson [Wed, 18 Sep 2013 19:53:09 +0000 (12:53 -0700)] 
tcg: Change tcg_global_mem_new_* to take a TCGv_ptr

Thus, use cpu_env as the parameter, not TCG_AREG0 directly.
Update all uses in the translators.

Reviewed-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
5 years agotcg: Remove lingering references to gen_opc_buf
Richard Henderson [Fri, 11 Dec 2015 17:17:45 +0000 (09:17 -0800)] 
tcg: Remove lingering references to gen_opc_buf

Three in comments and one in code in the stub tcg_liveness_analysis.

Signed-off-by: Richard Henderson <rth@twiddle.net>
5 years agotcg: Respect highwater in tcg_out_tb_finalize
Richard Henderson [Wed, 2 Dec 2015 21:59:59 +0000 (13:59 -0800)] 
tcg: Respect highwater in tcg_out_tb_finalize

Undo the workaround at b17a6d3390f87620735f7efb03bb1c96682ff449.

If there are lots of memory operations in a TB, the slow path code
can exceed the highwater reservation.  Add a check within the loop.

Tested-by: Aurelien Jarno <aurelien@aurel32.net>
Reviewed-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
5 years agoqapi: Add missing JSON files in build dependencies
Lluís Vilanova [Thu, 4 Feb 2016 18:29:17 +0000 (19:29 +0100)] 
qapi: Add missing JSON files in build dependencies

Forgotten in commit 1dde0f4 (trace.json) and commit fafa4d5
(rocker.json).

Signed-off-by: Lluís Vilanova <vilanova@ac.upc.edu>
Message-Id: <145461055662.15201.2702170180078718114.stgit@localhost>
Reviewed-by: Eric Blake <eblake@redhat.com>
[Commit message tweaked]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqapi: Fix compilation failure on MIPS and SPARC
Eric Blake [Tue, 2 Feb 2016 14:51:41 +0000 (07:51 -0700)] 
qapi: Fix compilation failure on MIPS and SPARC

Commit 86f4b687 broke compilation on MIPS and SPARC, which have a
preprocessor pollution of '#define mips 1' and '#define sparc 1',
respectively.  Treat it the same way as we do for the pollution with
'unix', so that QMP remains backwards compatible and only the C code
needs to use the alternative 'q_mips', 'q_sparc' spelling.

CC: James Hogan <james.hogan@imgtec.com>
CC: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Eric Blake <eblake@redhat.com>
Tested-by: James Hogan <james.hogan@imgtec.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqmp: Don't abuse stack to track qmp-output root
Eric Blake [Fri, 29 Jan 2016 13:49:01 +0000 (06:49 -0700)] 
qmp: Don't abuse stack to track qmp-output root

The previous commit documented an inconsistency in how we are
using the stack of qmp-output-visitor.  Normally, pushing a
single top-level object puts the object on the stack twice:
once as the root, and once as the current container being
appended to; but popping that struct only pops once.  However,
qmp_ouput_add() was trying to either set up the added object
as the new root (works if you parse two top-level scalars in a
row: the second replaces the first as the root) or as a member
of the current container (works as long as you have an open
container on the stack; but if you have popped the first
top-level container, it then resolves to the root and still
tries to add into that existing container).

Fix the stupidity by not tracking two separate things in the
stack.  Drop the now-useless qmp_output_first() and
qmp_output_last() while at it.

Saved for a later patch: we still are rather sloppy in that
qmp_output_get_object() can be called in the middle of a parse,
rather than requiring that a visit is complete.

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1454075341-13658-26-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqmp: Fix reference-counting of qnull on empty output visit
Eric Blake [Fri, 29 Jan 2016 13:49:00 +0000 (06:49 -0700)] 
qmp: Fix reference-counting of qnull on empty output visit

Commit 6c2f9a15 ensured that we would not return NULL when the
caller used an output visitor but had nothing to visit. But
in doing so, it added a FIXME about a reference count leak
that could abort qemu in the (unlikely) case of SIZE_MAX such
visits (more plausible on 32-bit).  (Although that commit
suggested we might fix it in time for 2.5, we ran out of time;
fortunately, it is unlikely enough to bite that it was not
worth worrying about during the 2.5 release.)

This fixes things by documenting the internal contracts, and
explaining why the internal function can return NULL and only
the public facing interface needs to worry about qnull(),
thus avoiding over-referencing the qnull_ global object.

It does not, however, fix the stupidity of the stack mixing
up two separate pieces of information; add a FIXME to explain
that issue, which will be fixed shortly in a future patch.

Signed-off-by: Eric Blake <eblake@redhat.com>
Cc: qemu-stable@nongnu.org
Message-Id: <1454075341-13658-25-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqapi: Drop unused error argument for list and implicit struct
Eric Blake [Fri, 29 Jan 2016 13:48:59 +0000 (06:48 -0700)] 
qapi: Drop unused error argument for list and implicit struct

No backend was setting an error when ending the visit of a list or
implicit struct, or when moving to the next list node.  Make the
callers a bit easier to follow by making this a part of the contract,
and removing the errp argument - callers can then unconditionally end
an object as part of cleanup without having to think about whether a
second error is dominated by a first, because there is no second
error.

A later patch will then tackle the larger task of splitting
visit_end_struct(), which can indeed set an error.

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1454075341-13658-24-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqapi: Tighten qmp_input_end_list()
Eric Blake [Fri, 29 Jan 2016 13:48:58 +0000 (06:48 -0700)] 
qapi: Tighten qmp_input_end_list()

The only way that qmp_input_pop() will set errp is if a dictionary
was the most recent thing pushed.  Since we don't have any
push(struct)/pop(list) or push(list)/pop(struct) mismatches (such
a mismatch is a programming bug), we therefore cannot set errp
inside qmp_input_end_list().  Make this obvious by
using &error_abort.  A later patch will then remove the errp
parameter of qmp_input_pop(), but that will first require the
larger task of splitting visit_end_struct().

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1454075341-13658-23-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqapi: Drop unused 'kind' for struct/enum visit
Eric Blake [Fri, 29 Jan 2016 13:48:57 +0000 (06:48 -0700)] 
qapi: Drop unused 'kind' for struct/enum visit

visit_start_struct() and visit_type_enum() had a 'kind' argument
that was usually set to either the stringized version of the
corresponding qapi type name, or to NULL (although some clients
didn't even get that right).  But nothing ever used the argument.
It's even hard to argue that it would be useful in a debugger,
as a stack backtrace also tells which type is being visited.

Therefore, drop the 'kind' argument as dead.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <1454075341-13658-22-git-send-email-eblake@redhat.com>
[Harmless rebase mistake cleaned up]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqapi: Swap 'name' in visit_* callbacks to match public API
Eric Blake [Fri, 29 Jan 2016 13:48:56 +0000 (06:48 -0700)] 
qapi: Swap 'name' in visit_* callbacks to match public API

As explained in the previous patches, matching argument order of
'name, &value' to JSON's "name":value makes sense.  However,
while the last two patches were easy with Coccinelle, I ended up
doing this one all by hand.  Now all the visitor callbacks match
the main interface.

The compiler is able to enforce that all clients match the changed
interface in visitor-impl.h, even where two pointers are being
swapped, because only one of the two pointers is const (if that
were not the case, then C's looseness on treating 'char *' like
'void *' would have made review a bit harder).

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <1454075341-13658-21-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqom: Swap 'name' next to visitor in ObjectPropertyAccessor
Eric Blake [Fri, 29 Jan 2016 13:48:55 +0000 (06:48 -0700)] 
qom: Swap 'name' next to visitor in ObjectPropertyAccessor

Similar to the previous patch, it's nice to have all functions
in the tree that involve a visitor and a name for conversion to
or from QAPI to consistently stick the 'name' parameter next
to the Visitor parameter.

Done by manually changing include/qom/object.h and qom/object.c,
then running this Coccinelle script and touching up the fallout
(Coccinelle insisted on adding some trailing whitespace).

    @ rule1 @
    identifier fn;
    typedef Object, Visitor, Error;
    identifier obj, v, opaque, name, errp;
    @@
     void fn
    - (Object *obj, Visitor *v, void *opaque, const char *name,
    + (Object *obj, Visitor *v, const char *name, void *opaque,
       Error **errp) { ... }

    @@
    identifier rule1.fn;
    expression obj, v, opaque, name, errp;
    @@
     fn(obj, v,
    -   opaque, name,
    +   name, opaque,
        errp)

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <1454075341-13658-20-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqapi: Swap visit_* arguments for consistent 'name' placement
Eric Blake [Fri, 29 Jan 2016 13:48:54 +0000 (06:48 -0700)] 
qapi: Swap visit_* arguments for consistent 'name' placement

JSON uses "name":value, but many of our visitor interfaces were
called with visit_type_FOO(v, &value, name, errp).  This can be
a bit confusing to have to mentally swap the parameter order to
match JSON order.  It's particularly bad for visit_start_struct(),
where the 'name' parameter is smack in the middle of the
otherwise-related group of 'obj, kind, size' parameters! It's
time to do a global swap of the parameter ordering, so that the
'name' parameter is always immediately after the Visitor argument.

Additional reason in favor of the swap: the existing include/qjson.h
prefers listing 'name' first in json_prop_*(), and I have plans to
unify that file with the qapi visitors; listing 'name' first in
qapi will minimize churn to the (admittedly few) qjson.h clients.

Later patches will then fix docs, object.h, visitor-impl.h, and
those clients to match.

Done by first patching scripts/qapi*.py by hand to make generated
files do what I want, then by running the following Coccinelle
script to affect the rest of the code base:
 $ spatch --sp-file script `git grep -l '\bvisit_' -- '**/*.[ch]'`
I then had to apply some touchups (Coccinelle insisted on TAB
indentation in visitor.h, and botched the signature of
visit_type_enum() by rewriting 'const char *const strings[]' to
the syntactically invalid 'const char*const[] strings').  The
movement of parameters is sufficient to provoke compiler errors
if any callers were missed.

    // Part 1: Swap declaration order
    @@
    type TV, TErr, TObj, T1, T2;
    identifier OBJ, ARG1, ARG2;
    @@
     void visit_start_struct
    -(TV v, TObj OBJ, T1 ARG1, const char *name, T2 ARG2, TErr errp)
    +(TV v, const char *name, TObj OBJ, T1 ARG1, T2 ARG2, TErr errp)
     { ... }

    @@
    type bool, TV, T1;
    identifier ARG1;
    @@
     bool visit_optional
    -(TV v, T1 ARG1, const char *name)
    +(TV v, const char *name, T1 ARG1)
     { ... }

    @@
    type TV, TErr, TObj, T1;
    identifier OBJ, ARG1;
    @@
     void visit_get_next_type
    -(TV v, TObj OBJ, T1 ARG1, const char *name, TErr errp)
    +(TV v, const char *name, TObj OBJ, T1 ARG1, TErr errp)
     { ... }

    @@
    type TV, TErr, TObj, T1, T2;
    identifier OBJ, ARG1, ARG2;
    @@
     void visit_type_enum
    -(TV v, TObj OBJ, T1 ARG1, T2 ARG2, const char *name, TErr errp)
    +(TV v, const char *name, TObj OBJ, T1 ARG1, T2 ARG2, TErr errp)
     { ... }

    @@
    type TV, TErr, TObj;
    identifier OBJ;
    identifier VISIT_TYPE =~ "^visit_type_";
    @@
     void VISIT_TYPE
    -(TV v, TObj OBJ, const char *name, TErr errp)
    +(TV v, const char *name, TObj OBJ, TErr errp)
     { ... }

    // Part 2: swap caller order
    @@
    expression V, NAME, OBJ, ARG1, ARG2, ERR;
    identifier VISIT_TYPE =~ "^visit_type_";
    @@
    (
    -visit_start_struct(V, OBJ, ARG1, NAME, ARG2, ERR)
    +visit_start_struct(V, NAME, OBJ, ARG1, ARG2, ERR)
    |
    -visit_optional(V, ARG1, NAME)
    +visit_optional(V, NAME, ARG1)
    |
    -visit_get_next_type(V, OBJ, ARG1, NAME, ERR)
    +visit_get_next_type(V, NAME, OBJ, ARG1, ERR)
    |
    -visit_type_enum(V, OBJ, ARG1, ARG2, NAME, ERR)
    +visit_type_enum(V, NAME, OBJ, ARG1, ARG2, ERR)
    |
    -VISIT_TYPE(V, OBJ, NAME, ERR)
    +VISIT_TYPE(V, NAME, OBJ, ERR)
    )

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <1454075341-13658-19-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqom: Use typedef for Visitor
Eric Blake [Fri, 29 Jan 2016 13:48:53 +0000 (06:48 -0700)] 
qom: Use typedef for Visitor

No need to repeat 'struct Visitor' when we already have it in
typedefs.h.  Omitting the redundant 'struct' also makes a later
patch easier to search for all object property callbacks that
are associated with a Visitor.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <1454075341-13658-18-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqapi: Don't cast Enum* to int*
Eric Blake [Fri, 29 Jan 2016 13:48:52 +0000 (06:48 -0700)] 
qapi: Don't cast Enum* to int*

C compilers are allowed to represent enums as a smaller type
than int, if all enum values fit in the smaller type.  There
are even compiler flags that force the use of this smaller
representation, although using them changes the ABI of a
binary. Therefore, our generated code for visit_type_ENUM()
(for all qapi enums) was wrong for casting Enum* to int* when
calling visit_type_enum().

It appears that no one has been using compiler ABI switches
for qemu, because if they had, we are potentially dereferencing
beyond bounds or even risking a SIGBUS on platforms where
unaligned pointer dereferencing is fatal.  But it is still
better to avoid the practice entirely, and just use the correct
types.

This matches the fix for alternate qapi types, done earlier in
commit 0426d53 "qapi: Simplify visiting of alternate types",
with generated code changing as:

| void visit_type_QType(Visitor *v, QType *obj, const char *name, Error **errp)
| {
|-    visit_type_enum(v, (int *)obj, QType_lookup, "QType", name, errp);
|+    int value = *obj;
|+    visit_type_enum(v, &value, QType_lookup, "QType", name, errp);
|+    *obj = value;
| }

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <1454075341-13658-17-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqapi: Consolidate visitor small integer callbacks
Eric Blake [Fri, 29 Jan 2016 13:48:51 +0000 (06:48 -0700)] 
qapi: Consolidate visitor small integer callbacks

Commit 4e27e819 introduced optional visitor callbacks for all
sorts of int types, but no visitor has supplied any of the
callbacks for sizes less than 64 bits.  In other words, the
generic implementation based on using type_[u]int64() followed
by bounds-checking works just fine. In the interest of
simplicity, it's easier to make the visitor callback interface
not have to worry about the other sizes.

Adding some helper functions minimizes the boilerplate required
to correct FIXMEs added earlier with regards to questionable
reuse of errp, particularly now that we can guarantee from a
single file audit that value is unchanged if an error is set.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <1454075341-13658-16-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqapi: Make all visitors supply uint64 callbacks
Eric Blake [Fri, 29 Jan 2016 13:48:50 +0000 (06:48 -0700)] 
qapi: Make all visitors supply uint64 callbacks

Our qapi visitor contract supports multiple integer visitors,
but left the type_uint64 visitor as optional (falling back on
type_int64); which in turn can lead to awkward behavior with
numbers larger than INT64_MAX (the user has to be aware of
twos complement, and deal with negatives).

This patch does not address the disparity in handling large
values as negatives.  It merely moves the fallback from uint64
to int64 from the visitor core to the visitors, where the issue
can actually be fixed, by implementing the missing type_uint64()
callbacks on top of the respective type_int64() callbacks, and
with a FIXME comment explaining why that's wrong.

With that done, we now have a type_uint64() callback in every
driver, so we can make it mandatory from the core.  And although
the type_int64() callback can cover the entire valid range of
type_uint{8,16,32} on valid user input, using type_uint64() to
avoid mixed signedness makes more sense.

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1454075341-13658-15-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqapi: Prefer type_int64 over type_int in visitors
Eric Blake [Fri, 29 Jan 2016 13:48:49 +0000 (06:48 -0700)] 
qapi: Prefer type_int64 over type_int in visitors

The qapi builtin type 'int' is basically shorthand for the type
'int64'.  In fact, since no visitor was providing the optional
type_int64() callback, visit_type_int64() was just always falling
back to type_int(), cementing the equivalence between the types.

However, some visitors are providing a type_uint64() callback.
For purposes of code consistency, it is nicer if all visitors
use the paired type_int64/type_uint64 names rather than the
mismatched type_int/type_uint64.  So this patch just renames
the signed int callbacks in place, dropping the type_int()
callback as redundant, and a later patch will focus on the
unsigned int callbacks.

Add some FIXMEs to questionable reuse of errp in code touched
by the rename, while at it (the reuse works as long as the
callbacks don't modify value when setting an error, but it's not
a good example to set) - a later patch will then fix those.

No change in functionality here, although further cleanups are
in the pipeline.

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1454075341-13658-14-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqapi-visit: Kill unused visit_end_union()
Eric Blake [Fri, 29 Jan 2016 13:48:48 +0000 (06:48 -0700)] 
qapi-visit: Kill unused visit_end_union()

The generated code can call visit_end_union() without having called
visit_start_union().  Example:

        if (!*obj) {
            goto out_obj;
        }
        visit_type_CpuInfoBase_fields(v, (CpuInfoBase **)obj, &err);
        if (err) {
            goto out_obj; // if we go from here...
        }
        if (!visit_start_union(v, !!(*obj)->u.data, &err) || err) {
            goto out_obj;
        }
        switch ((*obj)->arch) {
    [...]
        }
    out_obj:
        // ... then *obj is true, and ...
        error_propagate(errp, err);
        err = NULL;
        if (*obj) {
            // we end up here
            visit_end_union(v, !!(*obj)->u.data, &err);
        }
        error_propagate(errp, err);

Harmless only because no visitor implements end_union().  Clean it up
anyway, by deleting the function as useless.

Messed up since we have visit_end_union (commit cee2ded).

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <1453902888-20457-3-git-send-email-armbru@redhat.com>
[expand scope of patch to delete rather than repair]
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1454075341-13658-13-git-send-email-eblake@redhat.com>

5 years agoqapi: Track all failures between visit_start/stop
Eric Blake [Fri, 29 Jan 2016 13:48:47 +0000 (06:48 -0700)] 
qapi: Track all failures between visit_start/stop

Inside the generated code between visit_start_struct() and
visit_end_struct(), we were blindly setting the error into
the caller's errp parameter.  But a future patch to split
visit_end_struct() will require that we take action based
on whether an error has occurred, which requires us to track
all actions through a local err.  Rewrite the visits to be
more in line with the other generated calls.

Generated code changes look like:

|     visit_start_struct(v, (void **)obj, "Abort", name, sizeof(Abort), &err);
|-    if (!err) {
|-        if (*obj) {
|-            visit_type_Abort_fields(v, obj, errp);
|-        }
|-        visit_end_struct(v, &err);
|+    if (err) {
|+        goto out;
|     }
|+    if (!*obj) {
|+        goto out_obj;
|+    }
|+    visit_type_Abort_fields(v, obj, &err);
|+    error_propagate(errp, err);
|+    err = NULL;
|+out_obj:
|+    visit_end_struct(v, &err);
|+out:
|     error_propagate(errp, err);
| }

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1454075341-13658-12-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqapi: Improve generated event use of qapi visitor
Eric Blake [Fri, 29 Jan 2016 13:48:46 +0000 (06:48 -0700)] 
qapi: Improve generated event use of qapi visitor

All other successful clients of visit_start_struct() were paired
with an unconditional visit_end_struct(); but the generated
code for events was relying on qmp_output_visitor_cleanup() to
work on an incomplete visit.  Alter the code to guarantee that
the struct is completed, which will make a future patch to
split visit_end_struct() easier to reason about.  While at it,
drop some assertions and comments that are not present in other
uses of the qmp output visitor, and pass NULL rather than "" as
the 'kind' parameter (matching most other uses where obj is NULL).

The changes to the generated code look like:

|     qmp = qmp_event_build_dict("DEVICE_TRAY_MOVED");
|
|     qov = qmp_output_visitor_new();
|-    g_assert(qov);
|-
|     v = qmp_output_get_visitor(qov);
|-    g_assert(v);
|
|-    /* Fake visit, as if all members are under a structure */
|-    visit_start_struct(v, NULL, "", "DEVICE_TRAY_MOVED", 0, &err);
|+    visit_start_struct(v, NULL, NULL, "DEVICE_TRAY_MOVED", 0, &err);
|     if (err) {
|         goto out;
|     }
|     visit_type_str(v, (char **)&device, "device", &err);
|     if (err) {
|-        goto out;
|+        goto out_obj;
|     }
|     visit_type_bool(v, &tray_open, "tray-open", &err);
|     if (err) {
|-        goto out;
|+        goto out_obj;
|     }
|-    visit_end_struct(v, &err);
|+out_obj:
|+    visit_end_struct(v, err ? NULL : &err);
|     if (err) {
|         goto out;
|     }
|
|     obj = qmp_output_get_qobject(qov);
|-    g_assert(obj != NULL);
|+    g_assert(obj);
|
|     qdict_put_obj(qmp, "data", obj);
|     emit(QAPI_EVENT_DEVICE_TRAY_MOVED, qmp, &err);

Note that the 'goto out_obj' with no intervening code before the
label, as well as the construct of 'err ? NULL : &err', are both
a bit unusual but also temporary; they get fixed in a later patch
that splits visit_end_struct() to drop its errp parameter by moving
some checking before the label.  But until that time, this was the
simplest way to avoid the appearance of passing a possibly-set
error to visit_end_struct(), even though actual code inspection
shows that visit_end_struct() for a QMP output visitor will never
set an error.

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1454075341-13658-11-git-send-email-eblake@redhat.com>
[Commit message's code diff tweaked]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoballoon: Improve use of qapi visitor
Eric Blake [Fri, 29 Jan 2016 13:48:45 +0000 (06:48 -0700)] 
balloon: Improve use of qapi visitor

Rework the control flow of balloon_stats_get_all() to make it
easier for a later patch to split visit_end_struct().  Also
switch to the uint64 visitor to match the data type.

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1454075341-13658-10-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agovl: Ensure qapi visitor properly ends struct visit
Eric Blake [Fri, 29 Jan 2016 13:48:44 +0000 (06:48 -0700)] 
vl: Ensure qapi visitor properly ends struct visit

Guarantee that visit_end_struct() is called if
visit_start_struct() succeeded.  This matches the behavior of
most other uses of visitors, and is a step towards the possibility
of a future patch that adds and enforces some tighter semantics to
the visitor interface (namely, cleanup of the visitor would no
longer have to mop up as many leftovers from an aborted partial
visit).

The change to code here matches the flow of hmp.c:hmp_object_add();
a later patch will then further simplify the cleanup logic of both
places by refactoring visit_end_struct() to not require a second
local error object.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <1454075341-13658-9-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agohmp: Cache use of qapi visitor
Eric Blake [Fri, 29 Jan 2016 13:48:43 +0000 (06:48 -0700)] 
hmp: Cache use of qapi visitor

Cache the visitor in a local variable instead of repeatedly
calling the accessor.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <1454075341-13658-8-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agohmp: Drop pointless allocation during qapi visit
Eric Blake [Fri, 29 Jan 2016 13:48:42 +0000 (06:48 -0700)] 
hmp: Drop pointless allocation during qapi visit

The qapi visitor contract allows us to visit a virtual structure,
where we don't have any corresponding qapi struct.  Most such uses
pass NULL for @obj; but these two callers were passing a dummy
pointer, which then gets allocated to heap memory but then
immediately freed without use.  Clean this up to suppress unwanted
allocation, like we do elsewhere.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <1454075341-13658-7-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqapi: Drop dead parameter in gen_params()
Eric Blake [Fri, 29 Jan 2016 13:48:41 +0000 (06:48 -0700)] 
qapi: Drop dead parameter in gen_params()

Commit 5cdc8831 reworked gen_params() to be simpler, but forgot
to clean up a now-unused errp named argument.

No change to generated code.

Reported-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1454075341-13658-6-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqapi: Dealloc visitor does not need a type_size()
Eric Blake [Fri, 29 Jan 2016 13:48:40 +0000 (06:48 -0700)] 
qapi: Dealloc visitor does not need a type_size()

The intent of having the visitor type_size() callback differ
from type_uint64() is to allow special handling for sizes; the
visitor core gracefully falls back to type_uint64() if there is
no need for the distinction.  Since the dealloc visitor does
nothing for any of the int visits, drop the pointless size
handler.

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1454075341-13658-5-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqapi: Drop dead dealloc visitor variable
Eric Blake [Fri, 29 Jan 2016 13:48:39 +0000 (06:48 -0700)] 
qapi: Drop dead dealloc visitor variable

Commit 0b9d8542 added StackEntry.is_list_head, but forgot to
delete the now-unused QapiDeallocVisitor.is_list_head.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <1454075341-13658-4-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqapi: Avoid use of misnamed DO_UPCAST()
Eric Blake [Fri, 29 Jan 2016 13:48:38 +0000 (06:48 -0700)] 
qapi: Avoid use of misnamed DO_UPCAST()

The macro DO_UPCAST() is incorrectly named: it converts from a
parent class to a derived class (which is a downcast).  Better,
and more consistent with some of the other qapi visitors, is
to use the container_of() macro through a to_FOO() helper.  Names
like 'to_ov()' may be a bit short, but for a static helper it
doesn't hurt too much, and matches existing practice in files
like qmp-input-visitor.c.

Our current definition of container_of() is weaker than
DO_UPCAST(), in that it does not require the derived class to
have Visitor as its first member, but this does not hurt our
usage patterns in qapi visitors.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <1454075341-13658-3-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agoqobject: Document more shortcomings in our number handling
Eric Blake [Fri, 29 Jan 2016 13:48:37 +0000 (06:48 -0700)] 
qobject: Document more shortcomings in our number handling

We've already documented that our JSON parsing is locale dependent;
but we should also document that our JSON output has the same
problem.  Additionally, JSON requires finite values (you have to
upgrade to JSON5 to get support for Inf or NaN), and our output
truncates floating point numbers to the point of losing significant
precision that could cause the receiver to read a different value.

Sadly, this series is not going to be the one that addresses these
problems.

Fix some trailing whitespace I noticed in the vicinity.

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1454075341-13658-2-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
5 years agotests: Use Python 2.6 "except E as ..." syntax
Markus Armbruster [Fri, 18 Dec 2015 07:52:44 +0000 (08:52 +0100)] 
tests: Use Python 2.6 "except E as ..." syntax

PEP 8 calls for it, because it's forward compatible with Python 3.
Supported since Python 2.6, which we require (commit fec2103).

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <1450425164-24969-5-git-send-email-armbru@redhat.com>

5 years agoRevert "tracetool: use Python 2.4-compatible exception handling syntax"
Markus Armbruster [Fri, 18 Dec 2015 07:52:43 +0000 (08:52 +0100)] 
Revert "tracetool: use Python 2.4-compatible exception handling syntax"

This reverts commit 662da3854e3f490223373b40afdcfcc339d14aa5.

We require Python 2.6 now (commit fec2103).

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <1450425164-24969-4-git-send-email-armbru@redhat.com>

5 years agoscripts/qmp: Use Python 2.6 "except E as ..." syntax
Markus Armbruster [Fri, 18 Dec 2015 07:52:42 +0000 (08:52 +0100)] 
scripts/qmp: Use Python 2.6 "except E as ..." syntax

PEP 8 calls for it, because it's forward compatible with Python 3.
Supported since Python 2.6, which we require (commit fec2103).

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <1450425164-24969-3-git-send-email-armbru@redhat.com>

5 years agoqapi: Use Python 2.6 "except E as ..." syntax
Markus Armbruster [Fri, 18 Dec 2015 07:52:41 +0000 (08:52 +0100)] 
qapi: Use Python 2.6 "except E as ..." syntax

PEP 8 calls for it, because it's forward compatible with Python 3.
Supported since Python 2.6, which we require (commit fec2103).

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <1450425164-24969-2-git-send-email-armbru@redhat.com>

5 years agoui/cocoa.m: Include qemu/osdep.h
Peter Maydell [Fri, 29 Jan 2016 16:23:34 +0000 (16:23 +0000)] 
ui/cocoa.m: Include qemu/osdep.h

Include "qemu/osdep.h". (This is a manual commit equivalent
to what the clean-includes script would do, because that
script can't handle ObjectiveC source files.)

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1454084614-5365-1-git-send-email-peter.maydell@linaro.org

5 years agoMerge remote-tracking branch 'remotes/mst/tags/for_upstream' into staging
Peter Maydell [Mon, 8 Feb 2016 11:25:31 +0000 (11:25 +0000)] 
Merge remote-tracking branch 'remotes/mst/tags/for_upstream' into staging

pc and misc cleanups and fixes, virtio optimizations

Included here:
Refactoring and bugfix patches in PC/ACPI.
New commands for ipmi.
Virtio optimizations.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
# gpg: Signature made Sat 06 Feb 2016 18:44:26 GMT using RSA key ID D28D5469
# gpg: Good signature from "Michael S. Tsirkin <mst@kernel.org>"
# gpg:                 aka "Michael S. Tsirkin <mst@redhat.com>"

* remotes/mst/tags/for_upstream: (45 commits)
  net: set endianness on all backend devices
  fix MSI injection on Xen
  intel_iommu: large page support
  dimm: Correct type of MemoryHotplugState->base
  pc: set the OEM fields in the RSDT and the FADT from the SLIC
  acpi: add function to extract oem_id and oem_table_id from the user's SLIC
  acpi: expose oem_id and oem_table_id in build_rsdt()
  acpi: take oem_id in build_header(), optionally
  pc: Eliminate PcGuestInfo struct
  pc: Move APIC and NUMA data from PcGuestInfo to PCMachineState
  pc: Move PcGuestInfo.fw_cfg to PCMachineState
  pc: Remove PcGuestInfo.isapc_ram_fw field
  pc: Remove RAM size fields from PcGuestInfo
  pc: Remove compat fields from PcGuestInfo
  acpi: Don't save PcGuestInfo on AcpiBuildState
  acpi: Remove guest_info parameters from functions
  pc: Simplify xen_load_linux() signature
  pc: Simplify pc_memory_init() signature
  pc: Eliminate struct PcGuestInfoState
  pc: Move PcGuestInfo declaration to top of file
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
5 years agonet: set endianness on all backend devices
Laurent Vivier [Wed, 13 Jan 2016 19:26:25 +0000 (20:26 +0100)] 
net: set endianness on all backend devices

commit 5be7d9f1b1452613b95c6ba70b8d7ad3d0797991
       vhost-net: tell tap backend about the vnet endianness

makes vhost net to set the endianness of the device, but only for
the first device.

In case of multiqueue, we have multiple devices... This patch sets the
endianness for all the devices of the interface.

Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: Jason Wang <jasowang@redhat.com>
5 years agofix MSI injection on Xen
Stefano Stabellini [Wed, 13 Jan 2016 14:59:09 +0000 (14:59 +0000)] 
fix MSI injection on Xen

On Xen MSIs can be remapped into pirqs, which are a type of event
channels. It's mostly for the benefit of PCI passthrough devices, to
avoid the overhead of interacting with the emulated lapic.

However remapping interrupts and MSIs is also supported for emulated
devices, such as the e1000 and virtio-net.

When an interrupt or an MSI is remapped into a pirq, masking and
unmasking is done by masking and unmasking the event channel. The
masking bit on the PCI config space or MSI-X table should be ignored,
but it isn't at the moment.

As a consequence emulated devices which use MSI or MSI-X, such as
virtio-net, don't work properly (the guest doesn't receive any
notifications). The mechanism was working properly when xen_apic was
introduced, but I haven't narrowed down which commit in particular is
causing the regression.

Fix the issue by ignoring the masking bit for MSI and MSI-X which have
been remapped into pirqs.

Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agointel_iommu: large page support
Jason Wang [Thu, 14 Jan 2016 05:47:24 +0000 (00:47 -0500)] 
intel_iommu: large page support

Current intel_iommu only supports 4K page which may not be sufficient
to cover guest working set. This patch tries to enable 2M and 1G mapping
for intel_iommu. This is also useful for future device IOTLB
implementation to have a better hit rate.

Major work is adding a page mask field on IOTLB entry to make it
support large page. And also use the slpte level as key to do IOTLB
lookup. MAMV was increased to 18 to support direct invalidation for 1G
mapping.

Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agodimm: Correct type of MemoryHotplugState->base
David Gibson [Thu, 21 Jan 2016 01:37:51 +0000 (12:37 +1100)] 
dimm: Correct type of MemoryHotplugState->base

The 'base' field of MemoryHotplugState is ram_addr_t, which indicates that
it exists in the abstract address space of RAM regions.

However, the actual usage of this field indicates that it is a concrete
physical address (it's passed as an offset to memory_region_add_subgregion
for example).

So, correct its type to 'hwaddr'.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Acked-by: Eduardo Habkost <ehabkost@redhat.com>
5 years agopc: set the OEM fields in the RSDT and the FADT from the SLIC
Laszlo Ersek [Mon, 18 Jan 2016 14:12:13 +0000 (15:12 +0100)] 
pc: set the OEM fields in the RSDT and the FADT from the SLIC

The Microsoft spec about the SLIC and MSDM ACPI tables at
<http://go.microsoft.com/fwlink/p/?LinkId=234834> requires the OEM ID and
OEM Table ID fields to be consistent between the SLIC and the RSDT/XSDT.
That further affects the FADT, because a similar match between the FADT
and the RSDT/XSDT is required by the ACPI spec in general.

This patch wires up the previous three patches.

Cc: "Michael S. Tsirkin" <mst@redhat.com> (supporter:ACPI/SMBIOS)
Cc: Igor Mammedov <imammedo@redhat.com> (supporter:ACPI/SMBIOS)
Cc: Paolo Bonzini <pbonzini@redhat.com> (maintainer:X86)
Cc: Richard W.M. Jones <rjones@redhat.com>
Cc: Aleksei Kovura <alex3kov@zoho.com>
Cc: Michael Tokarev <mjt@tls.msk.ru>
Cc: Steven Newbury <steve@snewbury.org.uk>
RHBZ: https://bugzilla.redhat.com/show_bug.cgi?id=1248758
LP: https://bugs.launchpad.net/qemu/+bug/1533848
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Steven Newbury <steve@snewbury.org.uk>
5 years agoacpi: add function to extract oem_id and oem_table_id from the user's SLIC
Laszlo Ersek [Mon, 18 Jan 2016 14:12:12 +0000 (15:12 +0100)] 
acpi: add function to extract oem_id and oem_table_id from the user's SLIC

The acpi_get_slic_oem() function stores pointers to these fields in the
(first) SLIC table that the user passes in with the -acpitable switch.

Cc: "Michael S. Tsirkin" <mst@redhat.com> (supporter:ACPI/SMBIOS)
Cc: Igor Mammedov <imammedo@redhat.com> (supporter:ACPI/SMBIOS)
Cc: Richard W.M. Jones <rjones@redhat.com>
Cc: Aleksei Kovura <alex3kov@zoho.com>
Cc: Michael Tokarev <mjt@tls.msk.ru>
Cc: Steven Newbury <steve@snewbury.org.uk>
RHBZ: https://bugzilla.redhat.com/show_bug.cgi?id=1248758
LP: https://bugs.launchpad.net/qemu/+bug/1533848
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Steven Newbury <steve@snewbury.org.uk>
5 years agoacpi: expose oem_id and oem_table_id in build_rsdt()
Laszlo Ersek [Mon, 18 Jan 2016 14:12:11 +0000 (15:12 +0100)] 
acpi: expose oem_id and oem_table_id in build_rsdt()

Since build_rsdt() is implemented as common utility code (in
"hw/acpi/aml-build.c"), it should expose -- and forward -- the oem_id and
oem_table_id parameters between board code and the generic build_header()
function.

Cc: "Michael S. Tsirkin" <mst@redhat.com> (supporter:ACPI/SMBIOS)
Cc: Igor Mammedov <imammedo@redhat.com> (supporter:ACPI/SMBIOS)
Cc: Shannon Zhao <zhaoshenglong@huawei.com> (maintainer:ARM ACPI Subsystem)
Cc: Paolo Bonzini <pbonzini@redhat.com> (maintainer:X86)
Cc: Richard W.M. Jones <rjones@redhat.com>
Cc: Aleksei Kovura <alex3kov@zoho.com>
Cc: Michael Tokarev <mjt@tls.msk.ru>
Cc: Steven Newbury <steve@snewbury.org.uk>
RHBZ: https://bugzilla.redhat.com/show_bug.cgi?id=1248758
LP: https://bugs.launchpad.net/qemu/+bug/1533848
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Shannon Zhao <shannon.zhao@linaro.org>
5 years agoacpi: take oem_id in build_header(), optionally
Laszlo Ersek [Mon, 18 Jan 2016 14:12:10 +0000 (15:12 +0100)] 
acpi: take oem_id in build_header(), optionally

This patch is the continuation of commit 8870ca0e94f2 ("acpi: support
specified oem table id for build_header"). It will allow us to control the
OEM ID field too in the SDT header.

Cc: "Michael S. Tsirkin" <mst@redhat.com> (supporter:ACPI/SMBIOS)
Cc: Igor Mammedov <imammedo@redhat.com> (supporter:ACPI/SMBIOS)
Cc: Xiao Guangrong <guangrong.xiao@linux.intel.com> (maintainer:NVDIMM)
Cc: Shannon Zhao <zhaoshenglong@huawei.com> (maintainer:ARM ACPI Subsystem)
Cc: Paolo Bonzini <pbonzini@redhat.com> (maintainer:X86)
Cc: Richard W.M. Jones <rjones@redhat.com>
Cc: Aleksei Kovura <alex3kov@zoho.com>
Cc: Michael Tokarev <mjt@tls.msk.ru>
Cc: Steven Newbury <steve@snewbury.org.uk>
RHBZ: https://bugzilla.redhat.com/show_bug.cgi?id=1248758
LP: https://bugs.launchpad.net/qemu/+bug/1533848
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Shannon Zhao <shannon.zhao@linaro.org>
5 years agopc: Eliminate PcGuestInfo struct
Eduardo Habkost [Fri, 11 Dec 2015 18:42:33 +0000 (16:42 -0200)] 
pc: Eliminate PcGuestInfo struct

The struct is not used for anything, now.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agopc: Move APIC and NUMA data from PcGuestInfo to PCMachineState
Eduardo Habkost [Fri, 11 Dec 2015 18:42:32 +0000 (16:42 -0200)] 
pc: Move APIC and NUMA data from PcGuestInfo to PCMachineState

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Marcel Apfelbaum <marcel@redhat.com>
5 years agopc: Move PcGuestInfo.fw_cfg to PCMachineState
Eduardo Habkost [Fri, 11 Dec 2015 18:42:31 +0000 (16:42 -0200)] 
pc: Move PcGuestInfo.fw_cfg to PCMachineState

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Marcel Apfelbaum <marcel@redhat.com>
5 years agopc: Remove PcGuestInfo.isapc_ram_fw field
Eduardo Habkost [Fri, 11 Dec 2015 18:42:30 +0000 (16:42 -0200)] 
pc: Remove PcGuestInfo.isapc_ram_fw field

The code can use the PCMachineClass.pci_enabled field directly.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Marcel Apfelbaum <marcel@redhat.com>
5 years agopc: Remove RAM size fields from PcGuestInfo
Eduardo Habkost [Fri, 11 Dec 2015 18:42:29 +0000 (16:42 -0200)] 
pc: Remove RAM size fields from PcGuestInfo

The ACPI code can use the PCMachineState fields directly.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Marcel Apfelbaum <marcel@redhat.com>
5 years agopc: Remove compat fields from PcGuestInfo
Eduardo Habkost [Fri, 11 Dec 2015 18:42:28 +0000 (16:42 -0200)] 
pc: Remove compat fields from PcGuestInfo

Remove the fields: legacy_acpi_table_size, has_acpi_build,
has_reserved_memory, and rsdp_in_ram from PcGuestInfo, and let
the existing code use the PCMachineClass fields directly.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Marcel Apfelbaum <marcel@redhat.com>
5 years agoacpi: Don't save PcGuestInfo on AcpiBuildState
Eduardo Habkost [Fri, 11 Dec 2015 18:42:27 +0000 (16:42 -0200)] 
acpi: Don't save PcGuestInfo on AcpiBuildState

We don't need to save the pointer on AcpiBuildState, as it is not
used anymore.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Marcel Apfelbaum <marcel@redhat.com>
5 years agoacpi: Remove guest_info parameters from functions
Eduardo Habkost [Fri, 11 Dec 2015 18:42:26 +0000 (16:42 -0200)] 
acpi: Remove guest_info parameters from functions

We can use PC_MACHINE(qdev_get_machine())->acpi_guest_info to get
guest_info.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agopc: Simplify xen_load_linux() signature
Eduardo Habkost [Fri, 11 Dec 2015 18:42:25 +0000 (16:42 -0200)] 
pc: Simplify xen_load_linux() signature

We can get the PcGuestInfo struct directly from PCMachineState,
and the return value is not needed at all.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Marcel Apfelbaum <marcel@redhat.com>
5 years agopc: Simplify pc_memory_init() signature
Eduardo Habkost [Fri, 11 Dec 2015 18:42:24 +0000 (16:42 -0200)] 
pc: Simplify pc_memory_init() signature

We can get the PcGuestInfo struct directly from PCMachineState,
and the return value is not needed at all.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Marcel Apfelbaum <marcel@redhat.com>
5 years agopc: Eliminate struct PcGuestInfoState
Eduardo Habkost [Fri, 11 Dec 2015 18:42:23 +0000 (16:42 -0200)] 
pc: Eliminate struct PcGuestInfoState

Instead of allocating a new struct just for PcGuestInfo and the
mchine_done Notifier, place them inside PCMachineState.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Marcel Apfelbaum <marcel@redhat.com>
5 years agopc: Move PcGuestInfo declaration to top of file
Eduardo Habkost [Fri, 11 Dec 2015 18:42:22 +0000 (16:42 -0200)] 
pc: Move PcGuestInfo declaration to top of file

The struct will be used inside PCMachineState.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Marcel Apfelbaum <marcel@redhat.com>
5 years agoipmi: add ACPI power and GUID commands
Cédric Le Goater [Mon, 25 Jan 2016 14:07:34 +0000 (15:07 +0100)] 
ipmi: add ACPI power and GUID commands

>From the specs (20.8 Get Device GUID Command), the command needs to
return a GUID (Globally Unique ID), or UUID, that should never change
over the lifetime of the device. qemu_uuid looked like a good
candidate to start with but we could use a specific BMC property also
if needed.

Signed-off-by: Cédric Le Goater <clg@fr.ibm.com>
Acked-by: Corey Minyard <cminyard@mvista.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agoipmi: add GET_SYS_RESTART_CAUSE chassis command
Cédric Le Goater [Mon, 25 Jan 2016 14:07:33 +0000 (15:07 +0100)] 
ipmi: add GET_SYS_RESTART_CAUSE chassis command

This is a simulator. Just return an unknown cause (0).

Signed-off-by: Cédric Le Goater <clg@fr.ibm.com>
Acked-by: Corey Minyard <cminyard@mvista.com>
Reviewed-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agoipmi: add get and set SENSOR_TYPE commands
Cédric Le Goater [Mon, 25 Jan 2016 14:07:32 +0000 (15:07 +0100)] 
ipmi: add get and set SENSOR_TYPE commands

Signed-off-by: Cédric Le Goater <clg@fr.ibm.com>
Acked-by: Corey Minyard <cminyard@mvista.com>
Reviewed-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agoipmi: introduce a struct ipmi_sdr_compact
Cédric Le Goater [Mon, 25 Jan 2016 14:07:31 +0000 (15:07 +0100)] 
ipmi: introduce a struct ipmi_sdr_compact

Currently, sdr attributes are identified using byte offsets and this
can be a bit confusing.

This patch adds a struct ipmi_sdr_compact conforming to the IPMI specs
and replaces byte offsets with names. It also introduces and uses a
struct ipmi_sdr_header in sections of the code where no assumption is
made on the type of SDR. This leave rooms to potential usage of other
types in the future.

Signed-off-by: Cédric Le Goater <clg@fr.ibm.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agoipmi: fix SDR length value
Cédric Le Goater [Mon, 25 Jan 2016 14:07:30 +0000 (15:07 +0100)] 
ipmi: fix SDR length value

The IPMI BMC simulator populates the SDR table with a set of initial
SDRs. The length of each SDR is taken from the record itself (byte 4)
which does not include the size of the header. But, the full length
(header + data) is required by the sdr_add_entry() routine.

Signed-off-by: Cédric Le Goater <clg@fr.ibm.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agoipmi: cleanup error_report messages
Cédric Le Goater [Mon, 25 Jan 2016 14:07:29 +0000 (15:07 +0100)] 
ipmi: cleanup error_report messages

Signed-off-by: Cédric Le Goater <clg@fr.ibm.com>
Cc: Greg Kurz <gkurz@linux.vnet.ibm.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agoipmi: replace *_MAXCMD defines
Cédric Le Goater [Mon, 25 Jan 2016 14:07:28 +0000 (15:07 +0100)] 
ipmi: replace *_MAXCMD defines

ARRAY_SIZE() is simple to use and removes the need to pre-define
the size of the command arrays.

Signed-off-by: Cédric Le Goater <clg@fr.ibm.com>
Reviewed-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agoipmi: replace goto by a return statement
Cédric Le Goater [Mon, 25 Jan 2016 14:07:27 +0000 (15:07 +0100)] 
ipmi: replace goto by a return statement

Each routine using the IPMI_ADD_RSP_DATA, IPMI_CHECK_CMD_LEN or
IPMI_CHECK_RESERVATION macros needs to define a goto label 'out' to
handle hidden errors. Using directly a return statement has the same
effect and it removes the fact that 'out' needs to be defined.

The code exits in ipmi_sim_handle_command() are a little different
from the rest and a "possible" error in the macro IPMI_ADD_RSP_DATA is
handled before making use of it. This might be a bit excessive as a
minimum response len is currently 300 bytes and the patch checks that
at least 3 are available.

Signed-off-by: Cédric Le Goater <clg@fr.ibm.com>
Reviewed-by: Marcel Apfelbaum <marcel@redhat.com>
Reviewed-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Corey Minyard <cminyard@mvista.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agohw/pci: ensure that only PCI/PCIe bridges can be attached to pxb/pxb-pcie devices
Marcel Apfelbaum [Mon, 18 Jan 2016 15:27:26 +0000 (17:27 +0200)] 
hw/pci: ensure that only PCI/PCIe bridges can be attached to pxb/pxb-pcie devices

PCI devices can't be plugged directly into PCI extra root bridges
because their resources can't be computed by firmware before the ACPI
tables are loaded.

Signed-off-by: Marcel Apfelbaum <marcel@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agovhost-user-test: use correct ROM to speed up and avoid spurious failures
Paolo Bonzini [Thu, 4 Feb 2016 15:00:52 +0000 (16:00 +0100)] 
vhost-user-test: use correct ROM to speed up and avoid spurious failures

The mechanism to get the option ROM for virtio-net does not block the
PCI ROM from being loaded. Therefore, in vhost-user-test there are
two entries in the boot menu for the virtio-net card: one as an
embedded option ROM, one from the ROM BAR.

The embedded option ROM in vhost-user-test is the non-EFI-enabled,
while the ROM BAR has an EFI-enabled ROM. The two are compiled with
slightly different parameters, where only the old BIOS-only one doesn't
have a timeout for the "Press Ctrl-B" banner. When using a new
machine type, therefore, the vhost-user-test has to wait for the
EFI-enabled ROM's banner to go away. There are several ways to fix
this:

1) fix the ROMs to have the same configuration

2) add ",romfile=" to the -device line

3) remove --option-rom and add the ROM file name to the -device line

4) use an old machine type

This patch chooses 3. In addition, the file name was wrong because
qtest runs QEMU relative to the top build directory, not to the
x86_64-softmmu/ subdirectory, which is fixed too.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agohw/pxb: add pxb devices to the bridge category
Marcel Apfelbaum [Wed, 3 Feb 2016 11:56:10 +0000 (13:56 +0200)] 
hw/pxb: add pxb devices to the bridge category

Signed-off-by: Marcel Apfelbaum <marcel@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agovirtio: combine write of an entry into used ring
Vincenzo Maffione [Sun, 31 Jan 2016 10:29:06 +0000 (11:29 +0100)] 
virtio: combine write of an entry into used ring

Fill in an element of the used ring with a single combined access to the
guest physical memory, rather than using two separated accesses.
This reduces the overhead due to expensive address translation.

Signed-off-by: Vincenzo Maffione <v.maffione@gmail.com>
Message-Id: <e4a89a767a4a92cbb6bcc551e151487eb36e1722.1450218353.git.v.maffione@gmail.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agovirtio: read avail_idx from VQ only when necessary
Vincenzo Maffione [Sun, 31 Jan 2016 10:29:05 +0000 (11:29 +0100)] 
virtio: read avail_idx from VQ only when necessary

The virtqueue_pop() implementation needs to check if the avail ring
contains some pending buffers. To perform this check, it is not
always necessary to fetch the avail_idx in the VQ memory, which is
expensive. This patch introduces a shadow variable tracking avail_idx
and modifies virtio_queue_empty() to access avail_idx in physical
memory only when necessary.

Signed-off-by: Vincenzo Maffione <v.maffione@gmail.com>
Message-Id: <b617d6459902773d9f4ab843bfaca764f5af8eda.1450218353.git.v.maffione@gmail.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agovirtio: cache used_idx in a VirtQueue field
Vincenzo Maffione [Sun, 31 Jan 2016 10:29:04 +0000 (11:29 +0100)] 
virtio: cache used_idx in a VirtQueue field

Accessing used_idx in the VQ requires an expensive access to
guest physical memory. Before this patch, 3 accesses are normally
done for each pop/push/notify call. However, since the used_idx is
only written by us, we can track it in our internal data structure.

Signed-off-by: Vincenzo Maffione <v.maffione@gmail.com>
Message-Id: <3d062ec54e9a7bf9fb325c1fd693564951f2b319.1450218353.git.v.maffione@gmail.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agovirtio: combine the read of a descriptor
Paolo Bonzini [Sun, 31 Jan 2016 10:29:03 +0000 (11:29 +0100)] 
virtio: combine the read of a descriptor

Compared to vring, virtio has a performance penalty of 10%.  Fix it
by combining all the reads for a descriptor in a single address_space_read
call.  This also simplifies the code nicely.

Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agovring: slim down allocation of VirtQueueElements
Paolo Bonzini [Sun, 31 Jan 2016 10:29:02 +0000 (11:29 +0100)] 
vring: slim down allocation of VirtQueueElements

Build the addresses and s/g lists on the stack, and then copy them
to a VirtQueueElement that is just as big as required to contain this
particular s/g list.  The cost of the copy is minimal compared to that
of a large malloc.

Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agovirtio: slim down allocation of VirtQueueElements
Paolo Bonzini [Sun, 31 Jan 2016 10:29:01 +0000 (11:29 +0100)] 
virtio: slim down allocation of VirtQueueElements

Build the addresses and s/g lists on the stack, and then copy them
to a VirtQueueElement that is just as big as required to contain this
particular s/g list.  The cost of the copy is minimal compared to that
of a large malloc.

When virtqueue_map is used on the destination side of migration or on
loadvm, the iovecs have already been split at memory region boundary,
so we can just reuse the out_num/in_num we find in the file.

Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agovirtio: introduce virtqueue_alloc_element
Paolo Bonzini [Sun, 31 Jan 2016 10:29:00 +0000 (11:29 +0100)] 
virtio: introduce virtqueue_alloc_element

Allocate the arrays for in_addr/out_addr/in_sg/out_sg outside the
VirtQueueElement.  For now, virtqueue_pop and vring_pop keep
allocating a very large VirtQueueElement.

Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agovirtio: introduce qemu_get/put_virtqueue_element
Paolo Bonzini [Sun, 31 Jan 2016 10:28:59 +0000 (11:28 +0100)] 
virtio: introduce qemu_get/put_virtqueue_element

Move allocation to virtio functions also when loading/saving a
VirtQueueElement.  This will also let the load/save functions
keep backwards compatibility when the VirtQueueElement layout
is changed.

Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
5 years agovirtio: move allocation to virtqueue_pop/vring_pop
Paolo Bonzini [Thu, 4 Feb 2016 14:26:51 +0000 (16:26 +0200)] 
virtio: move allocation to virtqueue_pop/vring_pop

The return code of virtqueue_pop/vring_pop is unused except to check for
errors or 0.  We can thus easily move allocation inside the functions
and just return a pointer to the VirtQueueElement.

The advantage is that we will be able to allocate only the space that
is needed for the actual size of the s/g list instead of the full
VIRTQUEUE_MAX_SIZE items.  Currently VirtQueueElement takes about 48K
of memory, and this kind of allocation puts a lot of stress on malloc.
By cutting the size by two or three orders of magnitude, malloc can
use much more efficient algorithms.

The patch is pretty large, but changes to each device are testable
more or less independently.  Splitting it would mostly add churn.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
5 years agoMerge remote-tracking branch 'remotes/amit-migration/tags/migration-for-2.6-2' into...
Peter Maydell [Fri, 5 Feb 2016 14:20:46 +0000 (14:20 +0000)] 
Merge remote-tracking branch 'remotes/amit-migration/tags/migration-for-2.6-2' into staging

Migration pull req.

Small fixes, nothing major.

# gpg: Signature made Fri 05 Feb 2016 13:51:30 GMT using RSA key ID 854083B6
# gpg: Good signature from "Amit Shah <amit@amitshah.net>"
# gpg:                 aka "Amit Shah <amit@kernel.org>"
# gpg:                 aka "Amit Shah <amitshah@gmx.net>"

* remotes/amit-migration/tags/migration-for-2.6-2:
  migration: fix bad string passed to error_report()
  static checker: e1000-82540em got aliased to e1000
  migration: remove useless code.
  qmp-commands.hx: Document the missing options for migration capability commands
  qmp-commands.hx: Fix the missing options for migration parameters commands
  migration/ram: Fix some helper functions' parameter to use PageSearchStatus
  savevm: Split load vm state function qemu_loadvm_state
  migration: rename 'file' in MigrationState to 'to_dst_file'
  ram: Split host_from_stream_offset() into two helper functions

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
5 years agomigration: fix bad string passed to error_report()
Greg Kurz [Fri, 5 Feb 2016 08:33:26 +0000 (09:33 +0100)] 
migration: fix bad string passed to error_report()

state->name does not contain a terminating '\0' and you may get:

Machine type received is 'pseries-2.3y�?' and local is 'pseries-2.4'
load of migration failed: Invalid argument

Let's add a precision modifier to fix this.

Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Message-Id: <20160205083201.2201.76109.stgit@bahia.huguette.org>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
5 years agostatic checker: e1000-82540em got aliased to e1000
Amit Shah [Fri, 18 Dec 2015 05:35:47 +0000 (11:05 +0530)] 
static checker: e1000-82540em got aliased to e1000

Commit 8304402033e8dbe8e379017d51ed1dd8344f1dce changed the name of the
e1000-82540em device to e1000.  This was flagged:

   Section "e1000-82540em" does not exist in dest

Add the mapping to the changed section names dictionary so the checker
can proceed.

Signed-off-by: Amit Shah <amit.shah@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <7ccfe834c897142dceaa4da87c13b7059fa12aa8.1450416947.git.amit.shah@redhat.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
5 years agomigration: remove useless code.
Liang Li [Wed, 27 Jan 2016 06:11:05 +0000 (14:11 +0800)] 
migration: remove useless code.

Since 's->state' will be set in migrate_init(), there is no
need to set it before calling migrate_init(). The code and
the related comments can be removed.

Signed-off-by: Liang Li <liang.z.li@intel.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Message-Id: <1453875065-24326-1-git-send-email-liang.z.li@intel.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>