sdk/emulator/qemu.git
8 years agoqapi: Drop unused c_null()
Eric Blake [Thu, 17 Mar 2016 22:48:36 +0000 (16:48 -0600)]
qapi: Drop unused c_null()

Now that we are always bulk-initializing a QAPI C struct to 0
(whether by g_malloc0() or by 'Type arg = {0};'), we no longer
have any clients of c_null() in the generator for per-element
initialization.  This patch is easy enough to revert if we find
a use in the future, but in the present, get rid of the dead code.

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1458254921-17042-12-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
8 years agoqapi: Inline gen_visit_members() into lone caller
Eric Blake [Thu, 17 Mar 2016 22:48:35 +0000 (16:48 -0600)]
qapi: Inline gen_visit_members() into lone caller

Commit 82ca8e46 noticed that we had multiple implementations of
visiting every member of a struct, and consolidated it into
gen_visit_fields() (now gen_visit_members()) with enough
parameters to cater to slight differences between the clients.
But recent exposure of implicit types has meant that we are now
down to a single use of that method, so we can clean up the
unused conditionals and just inline it into the remaining
caller: gen_visit_object_members().

Likewise, gen_err_check() no longer needs optional parameters,
as the lone use of non-defaults was via gen_visit_members().

No change to generated code.

Suggested-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1458254921-17042-11-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
8 years agoqapi-commands: Inline single-use helpers of gen_marshal()
Eric Blake [Thu, 17 Mar 2016 22:48:34 +0000 (16:48 -0600)]
qapi-commands: Inline single-use helpers of gen_marshal()

Originally, gen_marshal_input_visit() (or gen_visitor_input_block()
before commit f1538019) was factored out to make it easy to do two
passes of a visit to each member of a (possibly-implicit) object,
without duplicating lots of code.  But after recent changes, those
visits now occupy a single line of emitted code, and the helper
method has become a series of conditionals both before and after
the one important line, making it rather awkward to see at a glance
what gets emitted on the first (parsing) or second (deallocation)
pass.  It's a lot easier to read the generator code if we just
inline both uses directly into gen_marshal(), without all the
conditionals.

Once we've done that, it's easy to notice that gen_marshal_vars()
is used only once, and inlining it too lets us consolidate some
mcgen() calls that used to be split across helpers.

gen_call() remains a single-use helper function, but it has
enough indentation and complexity that inlining it would hamper
legibility.

No change to generated output.  The fact that the diffstat shows
a net reduction in lines is an argument in favor of this cleanup.

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1458254921-17042-10-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
8 years agoqapi-commands: Utilize implicit struct visits
Eric Blake [Thu, 17 Mar 2016 22:48:33 +0000 (16:48 -0600)]
qapi-commands: Utilize implicit struct visits

Rather than generate inline per-member visits, take advantage
of the 'visit_type_FOO_members()' function for command
marshalling.  This is possible now that implicit structs can be
visited like any other.  Generate call arguments from a stack-
allocated struct, rather than a list of local variables:

|@@ -57,26 +57,15 @@ void qmp_marshal_add_fd(QDict *args, QOb
|     QmpInputVisitor *qiv = qmp_input_visitor_new_strict(QOBJECT(args));
|     QapiDeallocVisitor *qdv;
|     Visitor *v;
|-    bool has_fdset_id = false;
|-    int64_t fdset_id = 0;
|-    bool has_opaque = false;
|-    char *opaque = NULL;
|+    q_obj_add_fd_arg arg = {0};
|
|     v = qmp_input_get_visitor(qiv);
|-    if (visit_optional(v, "fdset-id", &has_fdset_id)) {
|-        visit_type_int(v, "fdset-id", &fdset_id, &err);
|-        if (err) {
|-            goto out;
|-        }
|-    }
|-    if (visit_optional(v, "opaque", &has_opaque)) {
|-        visit_type_str(v, "opaque", &opaque, &err);
|-        if (err) {
|-            goto out;
|-        }
|+    visit_type_q_obj_add_fd_arg_members(v, &arg, &err);
|+    if (err) {
|+        goto out;
|     }
|
|-    retval = qmp_add_fd(has_fdset_id, fdset_id, has_opaque, opaque, &err);
|+    retval = qmp_add_fd(arg.has_fdset_id, arg.fdset_id, arg.has_opaque, arg.opaque, &err);
|     if (err) {
|         goto out;
|     }
|@@ -88,12 +77,7 @@ out:
|     qmp_input_visitor_cleanup(qiv);
|     qdv = qapi_dealloc_visitor_new();
|     v = qapi_dealloc_get_visitor(qdv);
|-    if (visit_optional(v, "fdset-id", &has_fdset_id)) {
|-        visit_type_int(v, "fdset-id", &fdset_id, NULL);
|-    }
|-    if (visit_optional(v, "opaque", &has_opaque)) {
|-        visit_type_str(v, "opaque", &opaque, NULL);
|-    }
|+    visit_type_q_obj_add_fd_arg_members(v, &arg, NULL);
|     qapi_dealloc_visitor_cleanup(qdv);
| }

This also has the nice side effect of eliminating a chance of
collision between argument QMP names and local variables.

This patch also paves the way for some followup simplifications
in the generator, in subsequent patches.

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1458254921-17042-9-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
8 years agoqapi-event: Utilize implicit struct visits
Eric Blake [Thu, 17 Mar 2016 22:48:32 +0000 (16:48 -0600)]
qapi-event: Utilize implicit struct visits

Rather than generate inline per-member visits, take advantage
of the 'visit_type_FOO_members()' function for emitting events.
This is possible now that implicit structs can be visited like
any other.  Generated code shrinks accordingly; by initializing
a struct based on parameters, through a new gen_param_var()
helper, like:

|@@ -338,6 +250,9 @@ void qapi_event_send_block_job_error(con
|     QMPEventFuncEmit emit = qmp_event_get_func_emit();
|     QmpOutputVisitor *qov;
|     Visitor *v;
|+    q_obj_BLOCK_JOB_ERROR_arg param = {
|+        (char *)device, operation, action
|+    };
|
|     if (!emit) {
|         return;
@@ -351,19 +266,7 @@ void qapi_event_send_block_job_error(con
|     if (err) {
|         goto out;
|     }
|-    visit_type_str(v, "device", (char **)&device, &err);
|-    if (err) {
|-        goto out_obj;
|-    }
|-    visit_type_IoOperationType(v, "operation", &operation, &err);
|-    if (err) {
|-        goto out_obj;
|-    }
|-    visit_type_BlockErrorAction(v, "action", &action, &err);
|-    if (err) {
|-        goto out_obj;
|-    }
|-out_obj:
|+    visit_type_q_obj_BLOCK_JOB_ERROR_arg_members(v, &param, &err);
|     visit_end_struct(v, err ? NULL : &err);

Notice that the initialization of 'param' has to cast away const
(just as the old gen_visit_members() had to do): we can't change
the signature of the user function (which uses 'const char *'), but
have to assign it to a non-const QAPI object (which requires
'char *').

While touching this, document with a FIXME comment that there is
still a potential collision between QMP members and our choice of
local variable names within qapi_event_send_FOO().

This patch also paves the way for some followup simplifications
in the generator, in subsequent patches.

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1458254921-17042-8-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
8 years agoqapi-event: Drop qmp_output_get_qobject() null check
Eric Blake [Thu, 17 Mar 2016 22:48:31 +0000 (16:48 -0600)]
qapi-event: Drop qmp_output_get_qobject() null check

qmp_output_get_qobject() was changed never to return null some time
ago (in commit 6c2f9a15), but the qapi_event_send_FOO() functions
still check.  Clean that up:

|@@ -28,7 +28,6 @@ void qapi_event_send_acpi_device_ost(ACP
|     QMPEventFuncEmit emit;
|     QmpOutputVisitor *qov;
|     Visitor *v;
|-    QObject *obj;
|
|     emit = qmp_event_get_func_emit();
|     if (!emit) {
|@@ -54,10 +53,7 @@ out_obj:
|         goto out;
|     }
|
|-    obj = qmp_output_get_qobject(qov);
|-    g_assert(obj);
|-
|-    qdict_put_obj(qmp, "data", obj);
|+    qdict_put_obj(qmp, "data", qmp_output_get_qobject(qov));
|     emit(QAPI_EVENT_ACPI_DEVICE_OST, qmp, &err);
|
| out:

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1458254921-17042-7-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
8 years agoqapi: Emit implicit structs in generated C
Eric Blake [Thu, 17 Mar 2016 22:48:30 +0000 (16:48 -0600)]
qapi: Emit implicit structs in generated C

We already have several places that want to visit all the members
of an implicit object within a larger context (simple union variant,
event with anonymous data, command with anonymous arguments struct);
and will be adding another one soon (the ability to declare an
anonymous base for a flat union).  Having a C struct declared for
these implicit types, along with a visit_type_FOO_members() helper
function, will make for fewer special cases in our generator.

We do not, however, need qapi_free_FOO() or visit_type_FOO()
functions for implicit types, because they should not be used
directly outside of the generated code.  This is done by adding a
conditional in visit_object_type() for both qapi-types.py and
qapi-visit.py based on the object name.  The comparison of
"name.startswith('q_')" is a bit hacky (it's basically duplicating
what .is_implicit() already uses), but beats changing the signature
of the visit_object_type() callback to pass a new 'implicit' flag.
The hack should be temporary: we are considering adding a future
patch that consolidates the narrow visit_object_type(..., base,
local_members, variants) and visit_object_type_flat(...,
all_members, variants) [where different sets of information are
already broken out, and the QAPISchemaObjectType is no longer
available] into a broader visit_object_type(obj_type) [where the
visitor can query the needed fields from obj_type directly].

Also, now that we WANT to output C code for implicits, we no longer
need the visit_needed() filter, leaving 'q_empty' as the only object
still needing a special case.  Remember, 'q_empty' is the only
built-in generated object, which means that without a special case
it would be emitted in multiple files (the main qapi-types.h and in
qga-qapi-types.h) causing compilation failure due to redefinition.
But since it has no members, it's easier to just avoid an attempt to
visit that particular type; since gen_object() is called recursively,
we also prime the objects_seen set to cover any recursion into the
empty type.

The patch relies on the changed naming of implicit types in the
previous patch.  It is a bit unfortunate that the generated struct
names and visit_type_FOO_members() don't match normal naming
conventions, but it's not too bad, since they will only be used in
generated code.

The generated code grows substantially in size: the implicit
'-wrapper' types must be emitted in qapi-types.h before any union
can include an unboxed member of that type.  Arguably, the '-args'
types could be emitted in a private header for just qapi-visit.c
and qmp-marshal.c, rather than polluting qapi-types.h; but adding
complexity to the generator to split the output location according
to role doesn't seem worth the maintenance costs.

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1458254921-17042-6-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
8 years agoqapi: Adjust names of implicit types
Eric Blake [Thu, 17 Mar 2016 22:48:29 +0000 (16:48 -0600)]
qapi: Adjust names of implicit types

The original choice of ':obj-' as the prefix for implicit types
made it obvious that we weren't going to clash with any user-defined
names, which cannot contain ':'.  But now we want to create structs
for implicit types, to get rid of special cases in the generators,
and our use of ':' in implicit names needs a tweak to produce valid
C code.

We could transliterate ':' to '_', except that C99 mandates that
"identifiers that begin with an underscore are always reserved for
use as identifiers with file scope in both the ordinary and tag name
spaces".  So it's time to change our naming convention: we can
instead use the 'q_' prefix that we reserved for ourselves back in
commit 9fb081e0.  Technically, since we aren't planning on exposing
the empty type in generated code, we could keep the name ':empty',
but renaming it to 'q_empty' makes the check for startswith('q_')
cover all implicit types, whether or not code is generated for them.

As long as we don't declare 'empty' or 'obj' ticklish, it shouldn't
clash with c_name() prepending 'q_' to the user's ticklish names.

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1458254921-17042-5-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
8 years agoqapi: Make c_type() more OO-like
Eric Blake [Thu, 17 Mar 2016 22:48:28 +0000 (16:48 -0600)]
qapi: Make c_type() more OO-like

QAPISchemaType.c_type() is a bit awkward: it takes two optional
boolean flags is_param and is_unboxed, and they should never both
be True.

Add a new method for each of the flags, and drop the flags from
c_type().

Most callers pass no flags; they remain unchanged.

One caller passes is_param=True; call the new .c_param_type()
instead.

One caller passes is_unboxed=True, except for simple union types.
This is actually an ugly special case that will go away soon, so
until then, we now have to call either .c_type() or the new
.c_unboxed_type().  Tolerable in the interim.

It requires slightly more Python, but is arguably easier to read.

Suggested-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1458254921-17042-4-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
8 years agoqapi: Fix command with named empty argument type
Eric Blake [Thu, 17 Mar 2016 22:48:27 +0000 (16:48 -0600)]
qapi: Fix command with named empty argument type

The generator special-cased

 { 'command':'foo', 'data': {} }

to avoid emitting a visitor variable, but failed to see that

 { 'struct':'NamedEmptyType, 'data': {} }
 { 'command':'foo', 'data':'NamedEmptyType' }

needs the same treatment.  There, the generator happily generates a
visitor to get no arguments, and a visitor to destroy no arguments;
and the compiler isn't happy with that, as demonstrated by the updated
qapi-schema-test.json:

  tests/test-qmp-marshal.c: In function ‘qmp_marshal_user_def_cmd0’:
  tests/test-qmp-marshal.c:264:14: error: variable ‘v’ set but not used [-Werror=unused-but-set-variable]
       Visitor *v;
                ^

No change to generated code except for the testsuite addition.

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1458254921-17042-3-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
8 years agoqapi: Assert in places where variants are not handled
Eric Blake [Thu, 17 Mar 2016 22:48:26 +0000 (16:48 -0600)]
qapi: Assert in places where variants are not handled

We are getting closer to the point where we could use one union
as the base or variant type within another union type (as long
as there are no collisions between any possible combination of
member names allowed across all discriminator choices).  But
until we get to that point, it is worth asserting that variants
are not present in places where we are not prepared to handle
them: when exploding a type into a parameter list, we do not
expect variants.  The qapi.py code is already checking this,
via the older check_type() method; but someday we hope to get
rid of that and move checking into QAPISchema*.check().  The
two asserts added here make sure any refactoring still catches
problems, and makes it locally obvious why we can iterate over
only type.members without worrying about type.variants.

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1458254921-17042-2-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
8 years agoMerge remote-tracking branch 'remotes/kevin/tags/for-upstream' into staging
Peter Maydell [Thu, 17 Mar 2016 15:59:42 +0000 (15:59 +0000)]
Merge remote-tracking branch 'remotes/kevin/tags/for-upstream' into staging

Block layer patches

# gpg: Signature made Thu 17 Mar 2016 15:49:29 GMT using RSA key ID C88F2FD6
# gpg: Good signature from "Kevin Wolf <kwolf@redhat.com>"

* remotes/kevin/tags/for-upstream: (29 commits)
  iotests: Test QUORUM_REPORT_BAD in fifo mode
  quorum: Emit QUORUM_REPORT_BAD for reads in fifo mode
  block: Use blk_co_pwritev() in blk_co_write_zeroes()
  block: Use blk_aio_prwv() for aio_read/write/write_zeroes
  block: Use blk_prw() in blk_pread()/blk_pwrite()
  block: Use blk_co_pwritev() in blk_write_zeroes()
  block: Pull up blk_read_unthrottled() implementation
  block: Use blk_co_pwritev() for blk_write()
  block: Use blk_co_preadv() for blk_read()
  block: Use BdrvChild in BlockBackend
  block: Remove bdrv_states list
  block: Use bdrv_next() instead of bdrv_states
  block: Rewrite bdrv_next()
  block: Add blk_next_root_bs()
  block: Add bdrv_next_monitor_owned()
  block: Move some bdrv_*_all() functions to BB
  blockdev: Remove blk_hide_on_behalf_of_hmp_drive_del()
  blockdev: Split monitor reference from BB creation
  blockdev: Separate BB name management
  blockdev: Add list of all BlockBackends
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoMerge remote-tracking branch 'mreitz/tags/pull-block-for-kevin-2016-03-17-v2' into...
Kevin Wolf [Thu, 17 Mar 2016 15:48:49 +0000 (16:48 +0100)]
Merge remote-tracking branch 'mreitz/tags/pull-block-for-kevin-2016-03-17-v2' into queue-block

Two quorum patches for the block queue, v2.

# gpg: Signature made Thu Mar 17 16:44:11 2016 CET using RSA key ID E838ACAD
# gpg: Good signature from "Max Reitz <mreitz@redhat.com>"

* mreitz/tags/pull-block-for-kevin-2016-03-17-v2:
  iotests: Test QUORUM_REPORT_BAD in fifo mode
  quorum: Emit QUORUM_REPORT_BAD for reads in fifo mode

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoiotests: Test QUORUM_REPORT_BAD in fifo mode
Alberto Garcia [Tue, 15 Mar 2016 09:41:36 +0000 (11:41 +0200)]
iotests: Test QUORUM_REPORT_BAD in fifo mode

Signed-off-by: Alberto Garcia <berto@igalia.com>
Message-id: c0a8dbfdbe939520cda5f661af6f1cd7b6b4df9d.1458034554.git.berto@igalia.com
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
8 years agoquorum: Emit QUORUM_REPORT_BAD for reads in fifo mode
Alberto Garcia [Tue, 15 Mar 2016 09:41:35 +0000 (11:41 +0200)]
quorum: Emit QUORUM_REPORT_BAD for reads in fifo mode

If there's an I/O error in one of Quorum children then QEMU
should emit QUORUM_REPORT_BAD. However this is not working with
read-pattern=fifo. This patch fixes this problem.

Signed-off-by: Alberto Garcia <berto@igalia.com>
Message-id: d57e39e8d3e8564003a1e2aadbd29c97286eb2d2.1458034554.git.berto@igalia.com
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
8 years agoblock: Use blk_co_pwritev() in blk_co_write_zeroes()
Kevin Wolf [Tue, 8 Mar 2016 12:47:53 +0000 (13:47 +0100)]
block: Use blk_co_pwritev() in blk_co_write_zeroes()

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblock: Use blk_aio_prwv() for aio_read/write/write_zeroes
Kevin Wolf [Tue, 8 Mar 2016 12:47:52 +0000 (13:47 +0100)]
block: Use blk_aio_prwv() for aio_read/write/write_zeroes

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblock: Use blk_prw() in blk_pread()/blk_pwrite()
Kevin Wolf [Tue, 8 Mar 2016 12:47:51 +0000 (13:47 +0100)]
block: Use blk_prw() in blk_pread()/blk_pwrite()

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblock: Use blk_co_pwritev() in blk_write_zeroes()
Kevin Wolf [Tue, 8 Mar 2016 12:47:50 +0000 (13:47 +0100)]
block: Use blk_co_pwritev() in blk_write_zeroes()

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblock: Pull up blk_read_unthrottled() implementation
Kevin Wolf [Tue, 8 Mar 2016 12:47:49 +0000 (13:47 +0100)]
block: Pull up blk_read_unthrottled() implementation

Use blk_read(), so that it goes through blk_co_preadv() like all read
requests from the BB to the BDS.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblock: Use blk_co_pwritev() for blk_write()
Kevin Wolf [Tue, 8 Mar 2016 12:47:48 +0000 (13:47 +0100)]
block: Use blk_co_pwritev() for blk_write()

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblock: Use blk_co_preadv() for blk_read()
Kevin Wolf [Tue, 8 Mar 2016 12:47:47 +0000 (13:47 +0100)]
block: Use blk_co_preadv() for blk_read()

This patch introduces blk_co_preadv() as a central function on the
BlockBackend level that is supposed to handle all read requests from the
BB to its root BDS eventually.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblock: Use BdrvChild in BlockBackend
Kevin Wolf [Tue, 8 Mar 2016 12:47:46 +0000 (13:47 +0100)]
block: Use BdrvChild in BlockBackend

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblock: Remove bdrv_states list
Max Reitz [Wed, 16 Mar 2016 18:54:45 +0000 (19:54 +0100)]
block: Remove bdrv_states list

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblock: Use bdrv_next() instead of bdrv_states
Max Reitz [Wed, 16 Mar 2016 18:54:44 +0000 (19:54 +0100)]
block: Use bdrv_next() instead of bdrv_states

There is no point in manually iterating through the bdrv_states list
when there is bdrv_next().

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblock: Rewrite bdrv_next()
Max Reitz [Wed, 16 Mar 2016 18:54:43 +0000 (19:54 +0100)]
block: Rewrite bdrv_next()

Instead of using the bdrv_states list, iterate over all the
BlockDriverStates attached to BlockBackends, and over all the
monitor-owned BDSs afterwards (except for those attached to a BB).

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblock: Add blk_next_root_bs()
Max Reitz [Wed, 16 Mar 2016 18:54:42 +0000 (19:54 +0100)]
block: Add blk_next_root_bs()

This function iterates over all BDSs attached to a BB. We are going to
need it when rewriting bdrv_next() so it no longer uses bdrv_states.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblock: Add bdrv_next_monitor_owned()
Max Reitz [Wed, 16 Mar 2016 18:54:41 +0000 (19:54 +0100)]
block: Add bdrv_next_monitor_owned()

Add a function for iterating over all monitor-owned BlockDriverStates so
the generic block layer can do so.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblock: Move some bdrv_*_all() functions to BB
Max Reitz [Wed, 16 Mar 2016 18:54:40 +0000 (19:54 +0100)]
block: Move some bdrv_*_all() functions to BB

Move bdrv_commit_all() and bdrv_flush_all() to the BlockBackend level.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblockdev: Remove blk_hide_on_behalf_of_hmp_drive_del()
Max Reitz [Wed, 16 Mar 2016 18:54:39 +0000 (19:54 +0100)]
blockdev: Remove blk_hide_on_behalf_of_hmp_drive_del()

We can basically inline it in hmp_drive_del(); monitor_remove_blk() is
called already, so we just need to call bdrv_make_anon(), too.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblockdev: Split monitor reference from BB creation
Max Reitz [Wed, 16 Mar 2016 18:54:38 +0000 (19:54 +0100)]
blockdev: Split monitor reference from BB creation

Before this patch, blk_new() automatically assigned a name to the new
BlockBackend and considered it referenced by the monitor. This patch
removes the implicit monitor_add_blk() call from blk_new() (and
consequently the monitor_remove_blk() call from blk_delete(), too) and
thus blk_new() (and related functions) no longer take a BB name
argument.

In fact, there is only a single point where blk_new()/blk_new_open() is
called and the new BB is monitor-owned, and that is in blockdev_init().
Besides thus relieving us from having to invent names for all of the BBs
we use in qemu-img, this fixes a bug where qemu cannot create a new
image if there already is a monitor-owned BB named "image".

If a BB and its BDS tree are created in a single operation, as of this
patch the BDS tree will be created before the BB is given a name
(whereas it was the other way around before). This results in minor
change to the output of iotest 087, whose reference output is amended
accordingly.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblockdev: Separate BB name management
Max Reitz [Wed, 16 Mar 2016 18:54:37 +0000 (19:54 +0100)]
blockdev: Separate BB name management

Introduce separate functions (monitor_add_blk() and
monitor_remove_blk()) which set or unset a BB name. Since the name is
equivalent to the monitor's reference to a BB, adding a name the same as
declaring the BB to be monitor-owned and removing it revokes this
status, hence the function names.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblockdev: Add list of all BlockBackends
Max Reitz [Wed, 16 Mar 2016 18:54:36 +0000 (19:54 +0100)]
blockdev: Add list of all BlockBackends

While monitor_block_backends contains nearly all BBs, we sometimes
really need all BBs. To this end, this patch adds the block_backend
list.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblockdev: Rename blk_backends
Max Reitz [Wed, 16 Mar 2016 18:54:35 +0000 (19:54 +0100)]
blockdev: Rename blk_backends

The blk_backends list does not contain all BlockBackends but only the
ones which are referenced by the monitor, and that is not necessarily
true for every BlockBackend. Rename the list to monitor_block_backends
to make that fact clear.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblock: Drop BB name from bad option error
Max Reitz [Wed, 16 Mar 2016 18:54:34 +0000 (19:54 +0100)]
block: Drop BB name from bad option error

The information which BB is concerned does not seem useful enough to
justify its existence in most other place (which may be related to qemu
printing the -drive parameter in question anyway, and for blockdev-add
the attribution is naturally unambiguous). Furthermore, as of a future
patch, bdrv_get_device_name(bs) will always return the empty string
before bdrv_open_inherit() returns.

Therefore, just dropping that information seems to be the best course of
action.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoqapi: Drop QERR_UNKNOWN_BLOCK_FORMAT_FEATURE
Max Reitz [Wed, 16 Mar 2016 18:54:33 +0000 (19:54 +0100)]
qapi: Drop QERR_UNKNOWN_BLOCK_FORMAT_FEATURE

Just specifying a custom string is simpler in basically all places that
used it, and in addition, specifying the BB or node name is something we
generally do not do in other error messages when opening a BDS, so we
should not do it here.

This changes the output for iotest 036 (to the better, in my opinion),
so the reference output needs to be changed accordingly.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblock: Use blk_{commit,flush}_all() consistently
Max Reitz [Wed, 16 Mar 2016 18:54:32 +0000 (19:54 +0100)]
block: Use blk_{commit,flush}_all() consistently

Replace bdrv_commmit_all() and bdrv_flush_all() by their BlockBackend
equivalents.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblock: Add blk_commit_all()
Max Reitz [Wed, 16 Mar 2016 18:54:31 +0000 (19:54 +0100)]
block: Add blk_commit_all()

Later, we will remove bdrv_commit_all() and move its contents here, and
in order to replace bdrv_commit_all() calls by calls to blk_commit_all()
before doing so, we need to add it as an alias now.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblock: Use blk_next() in block-backend.c
Max Reitz [Wed, 16 Mar 2016 18:54:30 +0000 (19:54 +0100)]
block: Use blk_next() in block-backend.c

Instead of iterating directly through blk_backends, we can use
blk_next() instead. This gives us some abstraction from the list itself
which we can use to rename it, for example.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agomonitor: Use BB list for BB name completion
Max Reitz [Wed, 16 Mar 2016 18:54:29 +0000 (19:54 +0100)]
monitor: Use BB list for BB name completion

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8 years agoblock: Fix memory leak in hmp_drive_add_node()
Kevin Wolf [Wed, 16 Mar 2016 10:14:31 +0000 (11:14 +0100)]
block: Fix memory leak in hmp_drive_add_node()

hmp_drive_add_node() leaked qdict in the error path when no node-name is
specified.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
8 years agoblock: Fix qemu_root_bds_opts.head initialisation
Kevin Wolf [Tue, 15 Mar 2016 13:43:14 +0000 (14:43 +0100)]
block: Fix qemu_root_bds_opts.head initialisation

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
8 years agoMerge remote-tracking branch 'remotes/stefanha/tags/block-pull-request' into staging
Peter Maydell [Thu, 17 Mar 2016 11:27:54 +0000 (11:27 +0000)]
Merge remote-tracking branch 'remotes/stefanha/tags/block-pull-request' into staging

# gpg: Signature made Thu 17 Mar 2016 11:08:28 GMT using RSA key ID 81AB73C8
# gpg: Good signature from "Stefan Hajnoczi <stefanha@redhat.com>"
# gpg:                 aka "Stefan Hajnoczi <stefanha@gmail.com>"

* remotes/stefanha/tags/block-pull-request:
  Revert "qed: Implement .bdrv_drain"
  aio-posix: Change CONFIG_EPOLL to CONFIG_EPOLL_CREATE1

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoRevert "qed: Implement .bdrv_drain"
Stefan Hajnoczi [Tue, 8 Mar 2016 10:11:16 +0000 (10:11 +0000)]
Revert "qed: Implement .bdrv_drain"

This reverts commit df9a681dc9ad41c9cdeb9ecc5d060ba9abd27e01.

Note that commit df9a681dc9ad41c9cdeb9ecc5d060ba9abd27e01 included some
unrelated hunks, possibly due to a merge failure or an overlooked
squash.  This only reverts the qed .bdrv_drain() implementation.

The qed .bdrv_drain() implementation is unsafe and can lead to a double
request completion.

Paolo Bonzini reports:
"The problem is that bdrv_qed_drain calls qed_plug_allocating_write_reqs
unconditionally, but this is not correct if an allocating write is
queued.  In this case, qed_unplug_allocating_write_reqs will restart the
allocating write and possibly cause it to complete.  The aiocb however
is still in use for the L2/L1 table writes, and will then be completed
again as soon as the table writes are stable."

For QEMU 2.6 we can simply revert this commit.  A full solution for the
qed need check timer may be added if the bdrv_drain() implementation is
extended.

Reported-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Message-id: 1457431876-8475-1-git-send-email-stefanha@redhat.com

8 years agoaio-posix: Change CONFIG_EPOLL to CONFIG_EPOLL_CREATE1
Matthew Fortune [Tue, 23 Feb 2016 15:42:14 +0000 (15:42 +0000)]
aio-posix: Change CONFIG_EPOLL to CONFIG_EPOLL_CREATE1

CONFIG_EPOLL was being used to guard epoll_create1 which results
in build failures on CentOS 5.

Signed-off-by: Matthew Fortune <matthew.fortune@imgtec.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Message-id: 6D39441BF12EF246A7ABCE6654B023536BB85D08@hhmail02.hh.imgtec.org
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
8 years agoMerge remote-tracking branch 'remotes/ehabkost/tags/machine-pull-request' into staging
Peter Maydell [Thu, 17 Mar 2016 08:52:58 +0000 (08:52 +0000)]
Merge remote-tracking branch 'remotes/ehabkost/tags/machine-pull-request' into staging

Machine Core queue, 2016-03-16

# gpg: Signature made Wed 16 Mar 2016 18:57:34 GMT using RSA key ID 984DC5A6
# gpg: Good signature from "Eduardo Habkost <ehabkost@redhat.com>"

* remotes/ehabkost/tags/machine-pull-request:
  module: Rename machine_init() to opts_init()
  machine: Use type_init() to register machine classes

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agomodule: Rename machine_init() to opts_init()
Eduardo Habkost [Tue, 16 Feb 2016 20:59:07 +0000 (18:59 -0200)]
module: Rename machine_init() to opts_init()

The only remaining users of machine_init() only call
qemu_add_opts(). Rename machine_init() to opts_init() and move it
closer to the qemu_add_opts() calls on vl.c.

Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Igor Mammedov <imammedo@redhat.com>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Marcel Apfelbaum <marcel@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agomachine: Use type_init() to register machine classes
Eduardo Habkost [Tue, 16 Feb 2016 20:59:04 +0000 (18:59 -0200)]
machine: Use type_init() to register machine classes

Change all machine_init() users that simply call type_register*()
to use type_init().

Cc: Evgeny Voevodin <e.voevodin@samsung.com>
Cc: Maksim Kozlov <m.kozlov@samsung.com>
Cc: Igor Mitsyanko <i.mitsyanko@gmail.com>
Cc: Dmitry Solodkiy <d.solodkiy@samsung.com>
Cc: Peter Maydell <peter.maydell@linaro.org>
Cc: Rob Herring <robh@kernel.org>
Cc: Andrzej Zaborowski <balrogg@gmail.com>
Cc: Michael Walle <michael@walle.cc>
Cc: "Hervé Poussineau" <hpoussin@reactos.org>
Cc: Aurelien Jarno <aurelien@aurel32.net>
Cc: Leon Alrae <leon.alrae@imgtec.com>
Cc: Alexander Graf <agraf@suse.de>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Blue Swirl <blauwirbel@gmail.com>
Cc: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Acked-by: Marcel Apfelbaum <marcel@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
8 years agoMerge remote-tracking branch 'remotes/cody/tags/block-pull-request' into staging
Peter Maydell [Wed, 16 Mar 2016 18:20:10 +0000 (18:20 +0000)]
Merge remote-tracking branch 'remotes/cody/tags/block-pull-request' into staging

# gpg: Signature made Wed 16 Mar 2016 17:33:44 GMT using RSA key ID C0DE3057
# gpg: Good signature from "Jeffrey Cody <jcody@redhat.com>"
# gpg:                 aka "Jeffrey Cody <jeff@codyprime.org>"
# gpg:                 aka "Jeffrey Cody <codyprime@gmail.com>"

* remotes/cody/tags/block-pull-request:
  MAINTAINERS: Fix typo, block/stream.h -> block/stream.c
  block/sheepdog: fix argument passed to qemu_strtoul()

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoMerge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20160316-1' into...
Peter Maydell [Wed, 16 Mar 2016 17:43:37 +0000 (17:43 +0000)]
Merge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20160316-1' into staging

target-arm queue:
 * loader: Fix incorrect parameter name in load_image_mr()
 * Implement MRS (banked) and MSR (banked) instructions
 * virt: Implement versioning for machine model
 * i.MX: some initial patches preparing for i.MX6 support
 * new ASPEED AST2400 SoC and palmetto-bmc machine
 * bcm2835: add some more raspi2 devices
 * sd: fix segfault running "info qtree"

# gpg: Signature made Wed 16 Mar 2016 17:42:43 GMT using RSA key ID 14360CDE
# gpg: Good signature from "Peter Maydell <peter.maydell@linaro.org>"
# gpg:                 aka "Peter Maydell <pmaydell@gmail.com>"
# gpg:                 aka "Peter Maydell <pmaydell@chiark.greenend.org.uk>"

* remotes/pmaydell/tags/pull-target-arm-20160316-1: (21 commits)
  sd: Fix "info qtree" on boards with SD cards
  bcm2835_dma: add emulation of Raspberry Pi DMA controller
  bcm2835_property: implement framebuffer control/configuration properties
  bcm2835_fb: add framebuffer device for Raspberry Pi
  bcm2835_aux: add emulation of BCM2835 AUX (aka UART1) block
  bcm2835_peripherals: enable sdhci pending-insert quirk for raspberry pi
  hw/arm: Add palmetto-bmc machine
  hw/arm: Add ASPEED AST2400 SoC model
  hw/intc: Add (new) ASPEED VIC device model
  hw/timer: Add ASPEED timer device model
  i.MX: Add missing descriptions in devices.
  i.MX: Add i.MX6 CCM and ANALOG device.
  i.MX: Add the CLK_IPG_HIGH clock
  i.MX: Remove CCM useless clock computation handling.
  i.MX: Rename CCM NOCLK to CLK_NONE for naming consistency.
  i.MX: Allow GPT timer to rollover.
  arm: virt: Move machine class init code to the abstract machine type
  arm: virt: Add an abstract ARM virt machine type
  target-arm: Fix translation level on early translation faults
  target-arm: Implement MRS (banked) and MSR (banked) instructions
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agosd: Fix "info qtree" on boards with SD cards
Peter Maydell [Wed, 16 Mar 2016 17:06:02 +0000 (17:06 +0000)]
sd: Fix "info qtree" on boards with SD cards

The SD card object is not a SysBusDevice, so don't create it with
qdev_create() if we're not assigning it to a specific bus; use
object_new() instead.

This was causing 'info qtree' to segfault on boards with SD cards,
because qdev_create(NULL, TYPE_FOO) puts the created object on the
system bus, and then we may try to run functions like sysbus_dev_print()
on it, which fail when casting the object to SysBusDevice.

(This is the same mistake that we made with the NAND device
and fixed in commit 6749695eaaf346c1.)

Reported-by: xiaoqiang.zhao <zxq_yx_007@163.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: xiaoqiang.zhao <zxq_yx_007@163.com>
Message-id: 1458061009-7733-1-git-send-email-peter.maydell@linaro.org

8 years agobcm2835_dma: add emulation of Raspberry Pi DMA controller
Grégory ESTRADE [Wed, 16 Mar 2016 17:06:02 +0000 (17:06 +0000)]
bcm2835_dma: add emulation of Raspberry Pi DMA controller

At present, all DMA transfers complete inline (so a looping descriptor
queue will lock up the device). We also do not model pause/abort,
arbitrarion/priority, or debug features.

Signed-off-by: Grégory ESTRADE <gregory.estrade@gmail.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Andrew Baumann <Andrew.Baumann@microsoft.com>
Message-id: 1457467526-8840-6-git-send-email-Andrew.Baumann@microsoft.com
[AB: implement 2D mode, cleanup/refactoring for upstream submission]
Signed-off-by: Andrew Baumann <Andrew.Baumann@microsoft.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agobcm2835_property: implement framebuffer control/configuration properties
Grégory ESTRADE [Wed, 16 Mar 2016 17:06:01 +0000 (17:06 +0000)]
bcm2835_property: implement framebuffer control/configuration properties

The property channel driver now interfaces with the framebuffer device
to query and set framebuffer parameters. As a result of this, the "get
ARM RAM size" query now correctly returns the video RAM base address
(not total RAM size), and the ram-size property is no longer relevant
here.

Signed-off-by: Grégory ESTRADE <gregory.estrade@gmail.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Andrew Baumann <Andrew.Baumann@microsoft.com>
Message-id: 1457467526-8840-5-git-send-email-Andrew.Baumann@microsoft.com
[AB: cleanup/refactoring for upstream submission]
Signed-off-by: Andrew Baumann <Andrew.Baumann@microsoft.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agobcm2835_fb: add framebuffer device for Raspberry Pi
Grégory ESTRADE [Wed, 16 Mar 2016 17:06:01 +0000 (17:06 +0000)]
bcm2835_fb: add framebuffer device for Raspberry Pi

The framebuffer occupies the upper portion of memory (64MiB by
default), but it can only be controlled/configured via a system
mailbox or property channel (to be added by a subsequent patch).

Signed-off-by: Grégory ESTRADE <gregory.estrade@gmail.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Andrew Baumann <Andrew.Baumann@microsoft.com>
Message-id: 1457467526-8840-4-git-send-email-Andrew.Baumann@microsoft.com
[AB: added Windows (BGR) support and cleanup/refactoring for upstream submission]
Signed-off-by: Andrew Baumann <Andrew.Baumann@microsoft.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agobcm2835_aux: add emulation of BCM2835 AUX (aka UART1) block
Andrew Baumann [Wed, 16 Mar 2016 17:06:01 +0000 (17:06 +0000)]
bcm2835_aux: add emulation of BCM2835 AUX (aka UART1) block

At present only the core UART functions (data path for tx/rx) are
implemented, which is enough for UEFI to boot. The following
features/registers are unimplemented:
  * Line/modem control
  * Scratch register
  * Extra control
  * Baudrate
  * SPI interfaces

Signed-off-by: Andrew Baumann <Andrew.Baumann@microsoft.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1457467526-8840-3-git-send-email-Andrew.Baumann@microsoft.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agobcm2835_peripherals: enable sdhci pending-insert quirk for raspberry pi
Andrew Baumann [Wed, 16 Mar 2016 17:06:01 +0000 (17:06 +0000)]
bcm2835_peripherals: enable sdhci pending-insert quirk for raspberry pi

Signed-off-by: Andrew Baumann <Andrew.Baumann@microsoft.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1457467526-8840-2-git-send-email-Andrew.Baumann@microsoft.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agohw/arm: Add palmetto-bmc machine
Andrew Jeffery [Wed, 16 Mar 2016 17:06:01 +0000 (17:06 +0000)]
hw/arm: Add palmetto-bmc machine

The new machine is a thin layer over the AST2400 ARM926-based SoC[1].
Between the minimal machine and the current SoC implementation there is
enough functionality to boot an aspeed_defconfig Linux kernel to
userspace. Nothing yet is specific to the Palmetto's BMC (other than
using an AST2400 SoC), but creating specific machine types is preferable
to a generic machine that doesn't match any particular hardware.

[1] http://www.aspeedtech.com/products.php?fPath=20&rId=376

Signed-off-by: Andrew Jeffery <andrew@aj.id.au>
Message-id: 1458096317-25223-5-git-send-email-andrew@aj.id.au
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agohw/arm: Add ASPEED AST2400 SoC model
Andrew Jeffery [Wed, 16 Mar 2016 17:06:00 +0000 (17:06 +0000)]
hw/arm: Add ASPEED AST2400 SoC model

While the ASPEED AST2400 SoC[1] has a broad range of capabilities this
implementation is minimal, comprising an ARM926 processor, ASPEED VIC
and timer devices, and a 8250 UART.

[1] http://www.aspeedtech.com/products.php?fPath=20&rId=376

Signed-off-by: Andrew Jeffery <andrew@aj.id.au>
Message-id: 1458096317-25223-4-git-send-email-andrew@aj.id.au
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agohw/intc: Add (new) ASPEED VIC device model
Andrew Jeffery [Wed, 16 Mar 2016 17:06:00 +0000 (17:06 +0000)]
hw/intc: Add (new) ASPEED VIC device model

Implement a basic ASPEED VIC device model for the AST2400 SoC[1], with
enough functionality to boot an aspeed_defconfig Linux kernel. The model
implements the 'new' (revised) register set: While the hardware exposes
both the new and legacy register sets, accesses to the model's legacy
register set will not be serviced (however the access will be logged).

[1] http://www.aspeedtech.com/products.php?fPath=20&rId=376

Signed-off-by: Andrew Jeffery <andrew@aj.id.au>
Message-id: 1458096317-25223-3-git-send-email-andrew@aj.id.au
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agohw/timer: Add ASPEED timer device model
Andrew Jeffery [Wed, 16 Mar 2016 17:06:00 +0000 (17:06 +0000)]
hw/timer: Add ASPEED timer device model

Implement basic ASPEED timer functionality for the AST2400 SoC[1]: Up to
8 timers can independently be configured, enabled, reset and disabled.
Some hardware features are not implemented, namely clock value matching
and pulse generation, but the implementation is enough to boot the Linux
kernel configured with aspeed_defconfig.

[1] http://www.aspeedtech.com/products.php?fPath=20&rId=376

Signed-off-by: Andrew Jeffery <andrew@aj.id.au>
Message-id: 1458096317-25223-2-git-send-email-andrew@aj.id.au
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoi.MX: Add missing descriptions in devices.
Jean-Christophe Dubois [Wed, 16 Mar 2016 17:06:00 +0000 (17:06 +0000)]
i.MX: Add missing descriptions in devices.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Jean-Christophe Dubois <jcd@tribudubois.net>
Message-id: f1f565eb9dffdeb582feb1b15ba9e8b0afcf5468.1456868959.git.jcd@tribudubois.net
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoi.MX: Add i.MX6 CCM and ANALOG device.
Jean-Christophe Dubois [Wed, 16 Mar 2016 17:06:00 +0000 (17:06 +0000)]
i.MX: Add i.MX6 CCM and ANALOG device.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Jean-Christophe Dubois <jcd@tribudubois.net>
Message-id: 9fa80b4d8c5d0f50c94e77d74f952a7a665e168f.1456868959.git.jcd@tribudubois.net
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoi.MX: Add the CLK_IPG_HIGH clock
Jean-Christophe Dubois [Wed, 16 Mar 2016 17:06:00 +0000 (17:06 +0000)]
i.MX: Add the CLK_IPG_HIGH clock

EPIT, GPT and other i.MX timers are using "abstract" clocks among which
a CLK_IPG_HIGH clock.

On i.MX25 and i.MX31 CLK_IPG and CLK_IPG_HIGH are mapped to the same clock
but on other SOC like i.MX6 they are mapped to distinct clocks.

This patch add the CLK_IPG_HIGH to prepare for SOC where these 2 clocks are
different.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Jean-Christophe Dubois <jcd@tribudubois.net>
Message-id: 224bf650194760284cb40630e985867e1373276a.1456868959.git.jcd@tribudubois.net
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoi.MX: Remove CCM useless clock computation handling.
Jean-Christophe Dubois [Wed, 16 Mar 2016 17:05:59 +0000 (17:05 +0000)]
i.MX: Remove CCM useless clock computation handling.

Most clocks supported by the CCM are useless to the qemu framework.

Only clocks related to timers (EPIT, GPT, PWM, WATCHDOG, ...) are usefull
to QEMU code.

Therefore this patch removes clock computation handling for all clocks but:
* CLK_NONE,
* CLK_IPG,
* CLK_32k

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Jean-Christophe Dubois <jcd@tribudubois.net>
Message-id: 9e7222efb349801032e60c0f6b0fbad0e5dcf648.1456868959.git.jcd@tribudubois.net
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoi.MX: Rename CCM NOCLK to CLK_NONE for naming consistency.
Jean-Christophe Dubois [Wed, 16 Mar 2016 17:05:59 +0000 (17:05 +0000)]
i.MX: Rename CCM NOCLK to CLK_NONE for naming consistency.

This way all CCM clock defines/enums are named CLK_XXX

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Jean-Christophe Dubois <jcd@tribudubois.net>
Message-id: 8537df765c1713625c7a8b9aca4c7ca60b42e0c0.1456868959.git.jcd@tribudubois.net
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoi.MX: Allow GPT timer to rollover.
Jean-Christophe Dubois [Wed, 16 Mar 2016 17:05:59 +0000 (17:05 +0000)]
i.MX: Allow GPT timer to rollover.

GPT timer need to rollover when it reaches 0xffffffff.

It also need to reset to 0 when in "restart mode" and crossing the
compare 1 register.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Jean-Christophe Dubois <jcd@tribudubois.net>
Message-id: 6e2b36117a249a78bf822dd59a390368f407136e.1456868959.git.jcd@tribudubois.net
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoarm: virt: Move machine class init code to the abstract machine type
Wei Huang [Wed, 16 Mar 2016 17:05:59 +0000 (17:05 +0000)]
arm: virt: Move machine class init code to the abstract machine type

This patch moves the common class initialization code from
"virt-2.6" to the new abstract class. An empty property is added to
"virt-2.6" machine. In the meanwhile, related funtions are renamed
to "virt_2_6_*" for consistency.

Signed-off-by: Wei Huang <wei@redhat.com>
Message-id: 1457717778-17727-3-git-send-email-wei@redhat.com
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoarm: virt: Add an abstract ARM virt machine type
Wei Huang [Wed, 16 Mar 2016 17:05:59 +0000 (17:05 +0000)]
arm: virt: Add an abstract ARM virt machine type

In preparation for future ARM virt machine types, this patch creates
an abstract type for all ARM machines. The current machine type in
QEMU (i.e. "virt") is renamed to "virt-2.6", whose naming scheme is
similar to other architectures. For the purpose of backward compatibility,
"virt" is converted to an alias, pointing to "virt-2.6". With this patch,
"qemu -M ?" lists the following virtual machine types along with others:

virt                 QEMU 2.6 ARM Virtual Machine (alias of virt-2.6)
virt-2.6             QEMU 2.6 ARM Virtual Machine

Signed-off-by: Wei Huang <wei@redhat.com>
Message-id: 1457717778-17727-2-git-send-email-wei@redhat.com
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agotarget-arm: Fix translation level on early translation faults
Sergey Sorokin [Wed, 16 Mar 2016 17:05:58 +0000 (17:05 +0000)]
target-arm: Fix translation level on early translation faults

Qemu reports translation fault on 1st level instead of 0th level in case of
AArch64 address translation if the translation table walk is disabled or
the address is in the gap between the two regions.

Signed-off-by: Sergey Sorokin <afarallax@yandex.ru>
Message-id: 1457527503-25958-1-git-send-email-afarallax@yandex.ru
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoMAINTAINERS: Fix typo, block/stream.h -> block/stream.c
Jeff Cody [Thu, 10 Mar 2016 02:54:00 +0000 (21:54 -0500)]
MAINTAINERS: Fix typo, block/stream.h -> block/stream.c

There is no block/stream.h, the intended filename is block/stream.c
instead.

Signed-off-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Message-id: b9feeac95301c1b0b1c28a485da5e3781370c31a.1457578261.git.jcody@redhat.com

8 years agoblock/sheepdog: fix argument passed to qemu_strtoul()
Jeff Cody [Wed, 2 Mar 2016 16:24:42 +0000 (11:24 -0500)]
block/sheepdog: fix argument passed to qemu_strtoul()

The function qemu_strtoul() reads 'unsigned long' sized data,
which is larger than uint32_t on 64-bit machines.

Even though the snap_id field in the header is 32-bits, we must
accommodate the full size in qemu_strtoul().

This patch also adds more meaningful error handling to the
qemu_strtoul() call, and subsequent results.

Reported-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Hitoshi Mitake <mitake.hitoshi@lab.ntt.co.jp>
Message-id: e56fc50abedd9a112e0683342c8eafda063cd2f9.1456935548.git.jcody@redhat.com

8 years agotarget-arm: Implement MRS (banked) and MSR (banked) instructions
Peter Maydell [Wed, 16 Mar 2016 17:05:58 +0000 (17:05 +0000)]
target-arm: Implement MRS (banked) and MSR (banked) instructions

Starting with the ARMv7 Virtualization Extensions, the A32 and T32
instruction sets provide instructions "MSR (banked)" and "MRS
(banked)" which can be used to access registers for a mode other
than the current one:
 * R<m>_<mode>
 * ELR_hyp
 * SPSR_<mode>

Implement the missing instructions.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Acked-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Message-id: 1456762734-23939-1-git-send-email-peter.maydell@linaro.org

8 years agoloader: Fix incorrect parameter name in load_image_mr() macro
Jens Wiklander [Wed, 16 Mar 2016 17:05:58 +0000 (17:05 +0000)]
loader: Fix incorrect parameter name in load_image_mr() macro

Fix a typo in the load_image_mr() macro: 'mr' was written when
the parameter name is '_mr'. (This had no visible effects since
the single use of the macro used 'mr' as the argument.)

Fixes 76151cacfe956248a25b38b5e8429465584f47bb "loader: Add
load_image_mr() to load ROM image to a MemoryRegion"

Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoutil/base64.c: Clean includes
Peter Maydell [Tue, 23 Feb 2016 14:18:32 +0000 (14:18 +0000)]
util/base64.c: Clean includes

Remove unnecessary include of config-host.h.
(This was missed by the clean-includes script because of the
incorrect use of <> for a QEMU header.)

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-id: 1456237112-32662-5-git-send-email-peter.maydell@linaro.org

8 years agoupdate-linux-headers.sh: Fake types.h doesn't need to include anything
Peter Maydell [Tue, 23 Feb 2016 14:18:31 +0000 (14:18 +0000)]
update-linux-headers.sh: Fake types.h doesn't need to include anything

We have a fake linux/types.h which we create in update-linux-headers.h.
Now that every QEMU source file includes osdep.h, this fake header
doesn't need to include anything at all.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-id: 1456237112-32662-4-git-send-email-peter.maydell@linaro.org

8 years agoinclude/config.h: Remove
Peter Maydell [Tue, 23 Feb 2016 14:18:30 +0000 (14:18 +0000)]
include/config.h: Remove

include/config.h just includes config-target.h (and used to also
include config-host.h).
It is now obsolete and unused, because osdep.h does this job, so
remove it.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-id: 1456237112-32662-3-git-send-email-peter.maydell@linaro.org

8 years agoslirp/slirp.h: Remove now-empty #ifdefs
Peter Maydell [Tue, 23 Feb 2016 14:18:29 +0000 (14:18 +0000)]
slirp/slirp.h: Remove now-empty #ifdefs

After automatic cleanup to remove unnecessary #includes of headers that
osdep.h provides, slirp.h has a few now unnecessary #ifdef/#endif pairs;
remove them.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-id: 1456237112-32662-2-git-send-email-peter.maydell@linaro.org

8 years agoMerge remote-tracking branch 'remotes/armbru/tags/pull-error-2016-03-16' into staging
Peter Maydell [Wed, 16 Mar 2016 11:09:36 +0000 (11:09 +0000)]
Merge remote-tracking branch 'remotes/armbru/tags/pull-error-2016-03-16' into staging

Error reporting patches for 2016-03-16

# gpg: Signature made Wed 16 Mar 2016 09:57:00 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-error-2016-03-16:
  error: ensure errno detail is printed with error_abort

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoMerge remote-tracking branch 'remotes/armbru/tags/pull-monitor-2016-03-16' into staging
Peter Maydell [Wed, 16 Mar 2016 10:38:14 +0000 (10:38 +0000)]
Merge remote-tracking branch 'remotes/armbru/tags/pull-monitor-2016-03-16' into staging

Monitor patches for 2016-03-16

# gpg: Signature made Wed 16 Mar 2016 09:47:23 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-monitor-2016-03-16:
  qdev-monitor: add missing aliases for virtio device classes
  qdev-monitor: sort alias table by typename
  qdev-monitor: improve error message when alias device is unavailable

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoMerge remote-tracking branch 'remotes/dgibson/tags/ppc-for-2.6-20160316' into staging
Peter Maydell [Wed, 16 Mar 2016 10:09:26 +0000 (10:09 +0000)]
Merge remote-tracking branch 'remotes/dgibson/tags/ppc-for-2.6-20160316' into staging

ppc patch queue for 2016-03-16

Accumulated patches for target-ppc, pseries machine type and related
devices.  As we are now in soft freeze, these are mostly fixes.
   * Fix KVM migration for several SPRs that qemu didn't handle
   * Clean up handling of SDR1, which allows a fix to the gdbstub
   * Fix a race in spapr_rng
   * Fix a bug with multifunction hotplug

The exception is the 7 patches to allow EEH on spapr-pci-host-bridge
devices (rather than the special and poorly designed
spapr-vfio-pci-host-bridge device).  I believe these are low risk of
breaking non-EEH cases, and EEH cases were little used in practice
previously (since libvirt did not support the special device amongst
other things).  It did have a draft posted before the soft freeze,
removes a very ugly VFIO interface, and removes device we'd like to
deprecate sooner rather than later.  So, I'm hoping we can squeeze
these in during the soft freeze.

This includes two patches to the VFIO code, which Alex Williamson has
indicated he's ok with coming through my tree.

# gpg: Signature made Wed 16 Mar 2016 05:04:52 GMT using RSA key ID 20D9B392
# gpg: Good signature from "David Gibson <david@gibson.dropbear.id.au>"
# gpg:                 aka "David Gibson (Red Hat) <dgibson@redhat.com>"
# gpg:                 aka "David Gibson (ozlabs.org) <dgibson@ozlabs.org>"
# gpg: WARNING: This key is not certified with sufficiently trusted signatures!
# gpg:          It is not certain that the signature belongs to the owner.
# Primary key fingerprint: 75F4 6586 AE61 A66C C44E  87DC 6C38 CACA 20D9 B392

* remotes/dgibson/tags/ppc-for-2.6-20160316:
  vfio: Eliminate vfio_container_ioctl()
  spapr_pci: Remove finish_realize hook
  spapr_pci: (Mostly) remove spapr-pci-vfio-host-bridge
  spapr_pci: Allow EEH on spapr-pci-host-bridge
  spapr_pci: Eliminate class callbacks
  spapr_pci: Switch to vfio_eeh_as_op() interface
  vfio: Start improving VFIO/EEH interface
  spapr_rng: fix race with main loop
  target-ppc: Eliminate kvmppc_kern_htab global
  target-ppc: Add helpers for updating a CPU's SDR1 and external HPT
  target-ppc: Split out SREGS get/put functions
  spapr_pci: fix multifunction hotplug
  target-ppc: Add PVR for POWER8NVL processor
  ppc: Add a few more P8 PMU SPRs
  ppc: Fix migration of the TAR SPR
  ppc: Define the PSPB register on POWER8

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoerror: ensure errno detail is printed with error_abort
Daniel P. Berrange [Wed, 9 Mar 2016 17:28:24 +0000 (17:28 +0000)]
error: ensure errno detail is printed with error_abort

When &error_abort is passed in, the error reporting code
will print the current error message and then abort() the
process. Unfortunately at the time it aborts, we've not
yet appended the errno detail. This makes debugging certain
problems significantly harder as the log is incomplete.

Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Message-Id: <1457544504-8548-22-git-send-email-berrange@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
8 years agoMerge remote-tracking branch 'remotes/mst/tags/for_upstream' into staging
Peter Maydell [Wed, 16 Mar 2016 09:27:58 +0000 (09:27 +0000)]
Merge remote-tracking branch 'remotes/mst/tags/for_upstream' into staging

acpi: minor fix

Since previous pull acpi test triggers warnings,
fix it up.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
# gpg: Signature made Tue 15 Mar 2016 21:26:38 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:
  acpi-test: update UID for GSI links

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8 years agoqdev-monitor: add missing aliases for virtio device classes
Sascha Silbe [Thu, 18 Feb 2016 21:44:14 +0000 (22:44 +0100)]
qdev-monitor: add missing aliases for virtio device classes

virtio-{blk,balloon,net,serial} are aliases for their actual,
architecture-dependent implementations (*-ccw on s390x, *-pci on other
architectures supporting virtio). This makes it a lot easier to craft
qemu invocations that work on all supported architectures. Complete
the set to cover all existing non-abstract virtio device classes.

For virtio-balloon, only the CCW implementation was missing.

Signed-off-by: Sascha Silbe <silbe@linux.vnet.ibm.com>
Message-Id: <1455831854-49013-4-git-send-email-silbe@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
8 years agoqdev-monitor: sort alias table by typename
Sascha Silbe [Thu, 18 Feb 2016 21:44:13 +0000 (22:44 +0100)]
qdev-monitor: sort alias table by typename

Sort the alias table by typename so it's easier to see which aliases
exist.

Signed-off-by: Sascha Silbe <silbe@linux.vnet.ibm.com>
Message-Id: <1455831854-49013-3-git-send-email-silbe@linux.vnet.ibm.com>
Reviewed-by: Halil Pasic <pasic@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
8 years agoqdev-monitor: improve error message when alias device is unavailable
Sascha Silbe [Thu, 18 Feb 2016 21:44:12 +0000 (22:44 +0100)]
qdev-monitor: improve error message when alias device is unavailable

When trying to instantiate an alias that points to a device class that
doesn't exist, the error message looks like qemu misunderstood the
request:

$ s390x-softmmu/qemu-system-s390x -device virtio-gpu
qemu-system-s390x: -device virtio-gpu: 'virtio-gpu-ccw' is not a valid
device model name

Special-case the error message to make it explicit that alias
expansion is going on:

$ s390x-softmmu/qemu-system-s390x -device virtio-gpu
qemu-system-s390x: -device virtio-gpu: 'virtio-gpu' (alias
'virtio-gpu-ccw') is not a valid device model name

Suggested-By: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Sascha Silbe <silbe@linux.vnet.ibm.com>
Message-Id: <1455831854-49013-2-git-send-email-silbe@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
8 years agovfio: Eliminate vfio_container_ioctl()
David Gibson [Wed, 9 Mar 2016 00:57:20 +0000 (11:57 +1100)]
vfio: Eliminate vfio_container_ioctl()

vfio_container_ioctl() was a bad interface that bypassed abstraction
boundaries, had semantics that sat uneasily with its name, and was unsafe
in many realistic circumstances.  Now that spapr-pci-vfio-host-bridge has
been folded into spapr-pci-host-bridge, there are no more users, so remove
it.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Acked-by: Alex Williamson <alex.williamson@redhat.com>
8 years agospapr_pci: Remove finish_realize hook
David Gibson [Mon, 29 Feb 2016 06:20:00 +0000 (17:20 +1100)]
spapr_pci: Remove finish_realize hook

Now that spapr-pci-vfio-host-bridge is reduced to just a stub, there is
only one implementation of the finish_realize hook in sPAPRPHBClass.  So,
we can fold that implementation into its (single) caller, and remove the
hook.  That's the last thing left in sPAPRPHBClass, so that can go away as
well.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
8 years agospapr_pci: (Mostly) remove spapr-pci-vfio-host-bridge
David Gibson [Mon, 29 Feb 2016 06:19:50 +0000 (17:19 +1100)]
spapr_pci: (Mostly) remove spapr-pci-vfio-host-bridge

Now that the regular spapr-pci-host-bridge can handle EEH, there are only
two things that spapr-pci-vfio-host-bridge does differently:
    1. automatically sizes its DMA window to match the host IOMMU
    2. checks if the attached VFIO container is backed by the
       VFIO_SPAPR_TCE_IOMMU type on the host

(1) is not particularly useful, since the default window used by the
regular host bridge will work with the host IOMMU configuration on all
current systems anyway.

Plus, automatically changing guest visible configuration (such as the DMA
window) based on host settings is generally a bad idea.  It's not
definitively broken, since spapr-pci-vfio-host-bridge is only supposed to
support VFIO devices which can't be migrated anyway, but still.

(2) is not really useful, because if a guest tries to configure EEH on a
different host IOMMU, the first call will fail and that will be that.

It's possible there are scripts or tools out there which expect
spapr-pci-vfio-host-bridge, so we don't remove it entirely.  This patch
reduces it to just a stub for backwards compatibility.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
8 years agospapr_pci: Allow EEH on spapr-pci-host-bridge
David Gibson [Mon, 29 Feb 2016 06:19:42 +0000 (17:19 +1100)]
spapr_pci: Allow EEH on spapr-pci-host-bridge

Now that the EEH code is independent of the special
spapr-vfio-pci-host-bridge device, we can allow it on all spapr PCI
host bridges instead.  We do this by changing spapr_phb_eeh_available()
to be based on the vfio_eeh_as_ok() call instead of the host bridge class.

Because the value of vfio_eeh_as_ok() can change with devices being
hotplugged or unplugged, this can potentially lead to some strange edge
cases where the guest starts using EEH, then it starts failing because
of a change in status.

However, it's not really any worse than the current situation.  Cases that
would have worked previously will still work (i.e. VFIO devices from at
most one VFIO IOMMU group per vPHB), it's just that it's no longer
necessary to use spapr-vfio-pci-host-bridge with the groupid pre-specified.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
8 years agospapr_pci: Eliminate class callbacks
David Gibson [Mon, 29 Feb 2016 06:45:05 +0000 (17:45 +1100)]
spapr_pci: Eliminate class callbacks

The EEH operations in the spapr-vfio-pci-host-bridge no longer rely on the
special groupid field in sPAPRPHBVFIOState.  So we can simplify, removing
the class specific callbacks with direct calls based on a simple
spapr_phb_eeh_enabled() helper.  For now we implement that in terms of
a boolean in the class, but we'll continue to clean that up later.

On its own this is a rather strange way of doing things, but it's a useful
intermediate step to further cleanups.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
8 years agospapr_pci: Switch to vfio_eeh_as_op() interface
David Gibson [Mon, 29 Feb 2016 03:00:34 +0000 (14:00 +1100)]
spapr_pci: Switch to vfio_eeh_as_op() interface

This switches all EEH on VFIO operations in spapr_pci_vfio.c from the
broken vfio_container_ioctl() interface to the new vfio_as_eeh_op()
interface.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
8 years agovfio: Start improving VFIO/EEH interface
David Gibson [Wed, 9 Mar 2016 00:56:06 +0000 (11:56 +1100)]
vfio: Start improving VFIO/EEH interface

At present the code handling IBM's Enhanced Error Handling (EEH) interface
on VFIO devices operates by bypassing the usual VFIO logic with
vfio_container_ioctl().  That's a poorly designed interface with unclear
semantics about exactly what can be operated on.

In particular it operates on a single vfio container internally (hence the
name), but takes an address space and group id, from which it deduces the
container in a rather roundabout way.  groupids are something that code
outside vfio shouldn't even be aware of.

This patch creates new interfaces for EEH operations.  Internally we
have vfio_eeh_container_op() which takes a VFIOContainer object
directly.  For external use we have vfio_eeh_as_ok() which determines
if an AddressSpace is usable for EEH (at present this means it has a
single container with exactly one group attached), and vfio_eeh_as_op()
which will perform an operation on an AddressSpace in the unambiguous case,
and otherwise returns an error.

This interface still isn't great, but it's enough of an improvement to
allow a number of cleanups in other places.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Acked-by: Alex Williamson <alex.williamson@redhat.com>
8 years agospapr_rng: fix race with main loop
Greg Kurz [Fri, 11 Mar 2016 18:48:47 +0000 (19:48 +0100)]
spapr_rng: fix race with main loop

Since commit "60253ed1e6ec rng: add request queue support to rng-random",
the use of a spapr_rng device may hang vCPU threads.

The following path is taken without holding the lock to the main loop mutex:

h_random()
  rng_backend_request_entropy()
    rng_random_request_entropy()
      qemu_set_fd_handler()

The consequence is that entropy_available() may be called before the vCPU
thread could even queue the request: depending on the scheduling, it may
happen that entropy_available() does not call random_recv()->qemu_sem_post().
The vCPU thread will then sleep forever in h_random()->qemu_sem_wait().

This could not happen before 60253ed1e6ec because entropy_available() used
to call random_recv() unconditionally.

This patch ensures the lock is held to avoid the race.

Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Cédric Le Goater <clg@fr.ibm.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
8 years agotarget-ppc: Eliminate kvmppc_kern_htab global
David Gibson [Tue, 8 Mar 2016 00:35:15 +0000 (11:35 +1100)]
target-ppc: Eliminate kvmppc_kern_htab global

fa48b43 "target-ppc: Remove hack for ppc_hash64_load_hpte*() with HV KVM"
purports to remove a hack in the handling of hash page tables (HPTs)
managed by KVM instead of qemu.  However, it actually went in the wrong
direction.

That patch requires anything looking for an external HPT (that is one not
managed by the guest itself) to check both env->external_htab (for a qemu
managed HPT) and kvmppc_kern_htab (for a KVM managed HPT).  That's a
problem because kvmppc_kern_htab is local to mmu-hash64.c, but some places
which need to check for an external HPT are outside that, such as
kvm_arch_get_registers().  The latter was subtly broken by the earlier
patch such that gdbstub can no longer access memory.

Basically a KVM managed HPT is much more like a qemu managed HPT than it is
like a guest managed HPT, so the original "hack" was actually on the right
track.

This partially reverts fa48b43, so we again mark a KVM managed external HPT
by putting a special but non-NULL value in env->external_htab.  It then
goes further, using that marker to eliminate the kvmppc_kern_htab global
entirely.  The ppc_hash64_set_external_hpt() helper function is extended
to set that marker if passed a NULL value (if you're setting an external
HPT, but don't have an actual HPT to set, the assumption is that it must
be a KVM managed HPT).

This also has some flow-on changes to the HPT access helpers, required by
the above changes.

Reported-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Tested-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
8 years agotarget-ppc: Add helpers for updating a CPU's SDR1 and external HPT
David Gibson [Tue, 8 Mar 2016 00:33:46 +0000 (11:33 +1100)]
target-ppc: Add helpers for updating a CPU's SDR1 and external HPT

When a Power cpu with 64-bit hash MMU has it's hash page table (HPT)
pointer updated by a write to the SDR1 register we need to update some
derived variables.  Likewise, when the cpu is configured for an external
HPT (one not in the guest memory space) some derived variables need to be
updated.

Currently the logic for this is (partially) duplicated in ppc_store_sdr1()
and in spapr_cpu_reset().  In future we're going to need it in some other
places, so make some common helpers for this update.

In addition the new ppc_hash64_set_external_hpt() helper also updates
SDR1 in KVM - it's not updated by the normal runtime KVM <-> qemu CPU
synchronization.  In a sense this belongs logically in the
ppc_hash64_set_sdr1() helper, but that is called from
kvm_arch_get_registers() so can't itself call cpu_synchronize_state()
without infinite recursion.  In practice this doesn't matter because
the only other caller is TCG specific.

Currently there aren't situations where updating SDR1 at runtime in KVM
matters, but there are going to be in future.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
8 years agotarget-ppc: Split out SREGS get/put functions
David Gibson [Wed, 9 Mar 2016 00:58:33 +0000 (11:58 +1100)]
target-ppc: Split out SREGS get/put functions

Currently the getting and setting of Power MMU registers (sregs) take up
large inline chunks of the kvm_arch_get_registers() and
kvm_arch_put_registers() functions.  Especially since there are two
variants (for Book-E and Book-S CPUs), only one of which will be used in
practice, this is pretty hard to read.

This patch splits these out into helper functions for clarity.  No
functional change is expected.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
8 years agospapr_pci: fix multifunction hotplug
Michael Roth [Thu, 3 Mar 2016 21:55:36 +0000 (15:55 -0600)]
spapr_pci: fix multifunction hotplug

Since 3f1e147, QEMU has adopted a convention of supporting function
hotplug by deferring hotplug events until func 0 is hotplugged.
This is likely how management tools like libvirt would expose
such support going forward.

Since sPAPR guests rely on per-func events rather than
slot-based, our protocol has been to hotplug func 0 *first* to
avoid cases where devices appear within guests without func 0
present to avoid undefined behavior.

To remain compatible with new convention, defer hotplug in a
similar manner, but then generate events in 0-first order as we
did in the past. Once func 0 present, fail any attempts to plug
additional functions (as we do with PCIe).

For unplug, defer unplug operations in a similar manner, but
generate unplug events such that function 0 is removed last in guest.

Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
8 years agotarget-ppc: Add PVR for POWER8NVL processor
Alexey Kardashevskiy [Thu, 3 Mar 2016 00:08:19 +0000 (11:08 +1100)]
target-ppc: Add PVR for POWER8NVL processor

This adds a new POWER8+NVLink CPU PVR which core is identical to POWER8
but has a different PVR. The only available machine now has PVR
pvr 004c 0100 so this defines "POWER8NVL" alias as v1.0.

The corresponding kernel commit is
https://github.com/torvalds/linux/commit/ddee09c099c3
"powerpc: Add PVR for POWER8NVL processor"

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
8 years agoppc: Add a few more P8 PMU SPRs
Benjamin Herrenschmidt [Wed, 2 Mar 2016 20:19:22 +0000 (21:19 +0100)]
ppc: Add a few more P8 PMU SPRs

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
8 years agoppc: Fix migration of the TAR SPR
Thomas Huth [Wed, 2 Mar 2016 20:19:21 +0000 (21:19 +0100)]
ppc: Fix migration of the TAR SPR

The TAR special purpose register currently does not get migrated
under KVM because it does not get synchronized with the kernel.
Use spr_register_kvm() instead of spr_register() to fix this issue.

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