scripts: Coccinelle script to use ERRP_GUARD()

Script adds ERRP_GUARD() macro invocations where appropriate and
does corresponding changes in code (look for details in
include/qapi/error.h)

Usage example:
spatch --sp-file scripts/coccinelle/errp-guard.cocci \
 --macro-file scripts/cocci-macro-file.h --in-place --no-show-diff \
 --max-width 80 FILES...

Signed-off-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Reviewed-by: Markus Armbruster <[email protected]>
Signed-off-by: Markus Armbruster <[email protected]>
Message-Id: <20200707165037.1026246[email protected]>
Reviewed-by: Eric Blake <[email protected]>
[ERRP_AUTO_PROPAGATE() renamed to ERRP_GUARD(), and
auto-propagated-errp.cocci to errp-guard.cocci]

error: New macro ERRP_GUARD()

Introduce a new ERRP_GUARD() macro, to be used at start of functions
with an errp OUT parameter.

It has three goals:

1. Fix issue with error_fatal and error_prepend/error_append_hint: the
user can't see this additional information, because exit() happens in
error_setg earlier than information is added. [Reported by Greg Kurz]

2. Fix issue with error_abort and error_propagate: when we wrap
error_abort by local_err+error_propagate, the resulting coredump will
refer to error_propagate and not to the place where error happened.
(the macro itself doesn't fix the issue, but it allows us to [3.] drop
the local_err+error_propagate pattern, which will definitely fix the
issue) [Reported by Kevin Wolf]

3. Drop local_err+error_propagate pattern, which is used to workaround
void functions with errp parameter, when caller wants to know resulting
status. (Note: actually these functions could be merely updated to
return int error code).

To achieve these goals, later patches will add invocations
of this macro at the start of functions with either use
error_prepend/error_append_hint (solving 1) or which use
local_err+error_propagate to check errors, switching those
functions to use *errp instead (solving 2 and 3).

Signed-off-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Reviewed-by: Paul Durrant <[email protected]>
Reviewed-by: Greg Kurz <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
[Merge comments properly with recent commit "error: Document Error API
usage rules", and edit for clarity.  Put ERRP_AUTO_PROPAGATE() before
its helpers, and touch up style.  Tweak commit message.]
Signed-off-by: Markus Armbruster <[email protected]>
Message-Id: <20200707165037.1026246[email protected]>
[Rename ERRP_AUTO_PROPAGATE() to ERRP_GUARD(), tweak commit message
again]

hmp: Ignore Error objects where the return value suffices

qdev_print_props() receives and throws away Error objects just to
check for object_property_get_str() and object_property_print()
failure.  Unnecessary, both return suitable values, so use those
instead.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Message-Id: <20200707160613[email protected]>

qdev: Ignore Error objects where the return value suffices

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Message-Id: <20200707160613[email protected]>

qemu-img: Ignore Error objects where the return value suffices

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Message-Id: <20200707160613[email protected]>
[One more in img_amend() due to commit 0bc2a50e17 "qemu-option: Use
returned bool to check for failure"]

error: Avoid error_propagate() after migrate_add_blocker()

When migrate_add_blocker(blocker, &errp) is followed by
error_propagate(errp, err), we can often just as well do
migrate_add_blocker(..., errp).

Do that with this Coccinelle script:

    @@
    expression blocker, err, errp;
    expression ret;
    @@
    -    ret = migrate_add_blocker(blocker, &err);
    -    if (err) {
    +    ret = migrate_add_blocker(blocker, errp);
    +    if (ret < 0) {
             ... when != err;
    -        error_propagate(errp, err);
             ...
         }

    @@
    expression blocker, err, errp;
    @@
    -    migrate_add_blocker(blocker, &err);
    -    if (err) {
    +    if (migrate_add_blocker(blocker, errp) < 0) {
             ... when != err;
    -        error_propagate(errp, err);
             ...
         }

Double-check @err is not used afterwards.  Dereferencing it would be
use after free, but checking whether it's null would be legitimate.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>

qapi: Purge error_propagate() from QAPI core

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Message-Id: <20200707160613[email protected]>

qapi: Smooth visitor error checking in generated code

Use visitor functions' return values to check for failure.  Eliminate
error_propagate() that are now unnecessary.  Delete @err that are now
unused.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Message-Id: <20200707160613[email protected]>

qapi: Smooth another visitor error checking pattern

Convert

    visit_type_FOO(v, ..., &ptr, &err);
    ...
    if (err) {
        ...
    }

to

    visit_type_FOO(v, ..., &ptr, errp);
    ...
    if (!ptr) {
        ...
    }

for functions that set @ptr to non-null / null on success / error.

Eliminate error_propagate() that are now unnecessary.  Delete @err
that are now unused.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Message-Id: <20200707160613[email protected]>

block/parallels: Simplify parallels_open() after previous commit

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>

error: Reduce unnecessary error propagation

When all we do with an Error we receive into a local variable is
propagating to somewhere else, we can just as well receive it there
right away, even when we need to keep error_propagate() for other
error paths.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Message-Id: <20200707160613[email protected]>

error: Eliminate error_propagate() manually

When all we do with an Error we receive into a local variable is
propagating to somewhere else, we can just as well receive it there
right away.  The previous two commits did that for sufficiently simple
cases with Coccinelle.  Do it for several more manually.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Message-Id: <20200707160613[email protected]>

error: Eliminate error_propagate() with Coccinelle, part 2

When all we do with an Error we receive into a local variable is
propagating to somewhere else, we can just as well receive it there
right away.  The previous commit did that with a Coccinelle script I
consider fairly trustworthy.  This commit uses the same script with
the matching of return taken out, i.e. we convert

    if (!foo(..., &err)) {
        ...
        error_propagate(errp, err);
        ...
    }

to

    if (!foo(..., errp)) {
        ...
        ...
    }

This is unsound: @err could still be read between afterwards.  I don't
know how to express "no read of @err without an intervening write" in
Coccinelle.  Instead, I manually double-checked for uses of @err.

Suboptimal line breaks tweaked manually.  qdev_realize() simplified
further to placate scripts/checkpatch.pl.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Message-Id: <20200707160613[email protected]>

error: Eliminate error_propagate() with Coccinelle, part 1

When all we do with an Error we receive into a local variable is
propagating to somewhere else, we can just as well receive it there
right away.  Convert

    if (!foo(..., &err)) {
        ...
        error_propagate(errp, err);
        ...
        return ...
    }

to

    if (!foo(..., errp)) {
        ...
        ...
        return ...
    }

where nothing else needs @err.  Coccinelle script:

    @rule1 forall@
    identifier fun, err, errp, lbl;
    expression list args, args2;
    binary operator op;
    constant c1, c2;
    symbol false;
    @@
         if (
    (
    -        fun(args, &err, args2)
    +        fun(args, errp, args2)
    |
    -        !fun(args, &err, args2)
    +        !fun(args, errp, args2)
    |
    -        fun(args, &err, args2) op c1
    +        fun(args, errp, args2) op c1
    )
            )
         {
             ... when != err
                 when != lbl:
                 when strict
    -        error_propagate(errp, err);
             ... when != err
    (
             return;
    |
             return c2;
    |
             return false;
    )
         }

    @rule2 forall@
    identifier fun, err, errp, lbl;
    expression list args, args2;
    expression var;
    binary operator op;
    constant c1, c2;
    symbol false;
    @@
    -    var = fun(args, &err, args2);
    +    var = fun(args, errp, args2);
         ... when != err
         if (
    (
             var
    |
             !var
    |
             var op c1
    )
            )
         {
             ... when != err
                 when != lbl:
                 when strict
    -        error_propagate(errp, err);
             ... when != err
    (
             return;
    |
             return c2;
    |
             return false;
    |
             return var;
    )
         }

    @depends on rule1 || rule2@
    identifier err;
    @@
    -    Error *err = NULL;
         ... when != err

Not exactly elegant, I'm afraid.

The "when != lbl:" is necessary to avoid transforming

         if (fun(args, &err)) {
             goto out
         }
         ...
     out:
         error_propagate(errp, err);

even though other paths to label out still need the error_propagate().
For an actual example, see sclp_realize().

Without the "when strict", Coccinelle transforms vfio_msix_setup(),
incorrectly.  I don't know what exactly "when strict" does, only that
it helps here.

The match of return is narrower than what I want, but I can't figure
out how to express "return where the operand doesn't use @err".  For
an example where it's too narrow, see vfio_intx_enable().

Silently fails to convert hw/arm/armsse.c, because Coccinelle gets
confused by ARMSSE being used both as typedef and function-like macro
there.  Converted manually.

Line breaks tidied up manually.  One nested declaration of @local_err
deleted manually.  Preexisting unwanted blank line dropped in
hw/riscv/sifive_e.c.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Message-Id: <20200707160613[email protected]>

error: Avoid unnecessary error_propagate() after error_setg()

Replace

    error_setg(&err, ...);
    error_propagate(errp, err);

by

    error_setg(errp, ...);

Related pattern:

    if (...) {
        error_setg(&err, ...);
        goto out;
    }
    ...
 out:
    error_propagate(errp, err);
    return;

When all paths to label out are that way, replace by

    if (...) {
        error_setg(errp, ...);
        return;
    }

and delete the label along with the error_propagate().

When we have at most one other path that actually needs to propagate,
and maybe one at the end that where propagation is unnecessary, e.g.

    foo(..., &err);
    if (err) {
        goto out;
    }
    ...
    bar(..., &err);
 out:
    error_propagate(errp, err);
    return;

move the error_propagate() to where it's needed, like

    if (...) {
        foo(..., &err);
        error_propagate(errp, err);
        return;
    }
    ...
    bar(..., errp);
    return;

and transform the error_setg() as above.

In some places, the transformation results in obviously unnecessary
error_propagate().  The next few commits will eliminate them.

Bonus: the elimination of gotos will make later patches in this series
easier to review.

Candidates for conversion tracked down with this Coccinelle script:

    @@
    identifier err, errp;
    expression list args;
    @@
    -    error_setg(&err, args);
    +    error_setg(errp, args);
         ... when != err
         error_propagate(errp, err);

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Message-Id: <20200707160613[email protected]>

qdev: Use returned bool to check for failure, Coccinelle part

The previous commit enables conversion of

    qdev_prop_set_drive_err(..., &err);
    if (err) {
    ...
    }

to

    if (!qdev_prop_set_drive_err(..., errp)) {
    ...
    }

Coccinelle script:

    @@
    identifier fun = qdev_prop_set_drive_err;
    expression list args;
    typedef Error;
    Error *err;
    @@
    -    fun(args, &err);
    -    if (err)
    +    if (!fun(args, &err))
         {
             ...
         }

One line break tidied up manually.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Message-Id: <20200707160613[email protected]>

qdev: Make functions taking Error ** return bool, not void

See recent commit "error: Document Error API usage rules" for
rationale.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>

qom: Make functions taking Error ** return bool, not 0/-1

Just for consistency.  Also fix the example in object_set_props()'s
documentation.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>

qom: Use returned bool to check for failure, manual part

The previous commit used Coccinelle to convert from checking the Error
object to checking the return value.  Convert a few more manually.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>

qom: Use returned bool to check for failure, Coccinelle part

The previous commit enables conversion of

    foo(..., &err);
    if (err) {
        ...
    }

to

    if (!foo(..., errp)) {
        ...
    }

for QOM functions that now return true / false on success / error.
Coccinelle script:

    @@
    identifier fun = {
        object_apply_global_props, object_initialize_child_with_props,
        object_initialize_child_with_propsv, object_property_get,
        object_property_get_bool, object_property_parse, object_property_set,
        object_property_set_bool, object_property_set_int,
        object_property_set_link, object_property_set_qobject,
        object_property_set_str, object_property_set_uint, object_set_props,
        object_set_propv, user_creatable_add_dict,
        user_creatable_complete, user_creatable_del
    };
    expression list args, args2;
    typedef Error;
    Error *err;
    @@
    -    fun(args, &err, args2);
    -    if (err)
    +    if (!fun(args, &err, args2))
         {
             ...
         }

Fails to convert hw/arm/armsse.c, because Coccinelle gets confused by
ARMSSE being used both as typedef and function-like macro there.
Convert manually.

Line breaks tidied up manually.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>

qom: Make functions taking Error ** return bool, not void

See recent commit "error: Document Error API usage rules" for
rationale.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>

qom: Put name parameter before value / visitor parameter

The object_property_set_FOO() setters take property name and value in
an unusual order:

    void object_property_set_FOO(Object *obj, FOO_TYPE value,
                                 const char *name, Error **errp)

Having to pass value before name feels grating.  Swap them.

Same for object_property_set(), object_property_get(), and
object_property_parse().

Convert callers with this Coccinelle script:

    @@
    identifier fun = {
        object_property_get, object_property_parse, object_property_set_str,
        object_property_set_link, object_property_set_bool,
        object_property_set_int, object_property_set_uint, object_property_set,
        object_property_set_qobject
    };
    expression obj, v, name, errp;
    @@
    -    fun(obj, v, name, errp)
    +    fun(obj, name, v, errp)

Chokes on hw/arm/musicpal.c's lcd_refresh() with the unhelpful error
message "no position information".  Convert that one manually.

Fails to convert hw/arm/armsse.c, because Coccinelle gets confused by
ARMSSE being used both as typedef and function-like macro there.
Convert manually.

Fails to convert hw/rx/rx-gdbsim.c, because Coccinelle gets confused
by RXCPU being used both as typedef and function-like macro there.
Convert manually.  The other files using RXCPU that way don't need
conversion.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>
[Straightforwad conflict with commit 2336172d9b "audio: set default
value for pcspk.iobase property" resolved]

qom: Use return values to check for error where that's simpler

When using the Error object to check for error, we need to receive it
into a local variable, then propagate() it to @errp.

Using the return value permits allows receiving it straight to @errp.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>

qom: Don't handle impossible object_property_get_link() failure

Don't handle object_property_get_link() failure that can't happen
unless the programmer screwed up, pass &error_abort.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Message-Id: <20200707160613[email protected]>

qom: Crash more nicely on object_property_get_link() failure

Pass &error_abort instead of NULL where the returned value is
dereferenced or asserted to be non-null.  Drop a now redundant
assertion.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>

qom: Rename qdev_get_type() to object_get_type()

Commit 2f262e06f0 lifted qdev_get_type() from qdev to object without
renaming it accordingly.  Do that now.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>

qom: Use error_reportf_err() instead of g_printerr() in examples

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>

s390x/pci: Fix harmless mistake in zpci's property fid's setter

s390_pci_set_fid() sets zpci->fid_defined to true even when
visit_type_uint32() failed.  Reproducer: "-device zpci,fid=junk".
Harmless in practice, because qdev_device_add() then fails, throwing
away @zpci.  Fix it anyway.

Cc: Matthew Rosato <[email protected]>
Cc: Cornelia Huck <[email protected]>
Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Matthew Rosato <[email protected]>
Reviewed-by: Cornelia Huck <[email protected]>
Message-Id: <20200707160613[email protected]>

qapi: Use returned bool to check for failure, manual part

The previous commit used Coccinelle to convert from checking the Error
object to checking the return value.  Convert a few more manually.
Also tweak control flow in places to conform to the conventional "if
error bail out" pattern.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>

qapi: Use returned bool to check for failure, Coccinelle part

The previous commit enables conversion of

    visit_foo(..., &err);
    if (err) {
        ...
    }

to

    if (!visit_foo(..., errp)) {
        ...
    }

for visitor functions that now return true / false on success / error.
Coccinelle script:

    @@
    identifier fun =~ "check_list|input_type_enum|lv_start_struct|lv_type_bool|lv_type_int64|lv_type_str|lv_type_uint64|output_type_enum|parse_type_bool|parse_type_int64|parse_type_null|parse_type_number|parse_type_size|parse_type_str|parse_type_uint64|print_type_bool|print_type_int64|print_type_null|print_type_number|print_type_size|print_type_str|print_type_uint64|qapi_clone_start_alternate|qapi_clone_start_list|qapi_clone_start_struct|qapi_clone_type_bool|qapi_clone_type_int64|qapi_clone_type_null|qapi_clone_type_number|qapi_clone_type_str|qapi_clone_type_uint64|qapi_dealloc_start_list|qapi_dealloc_start_struct|qapi_dealloc_type_anything|qapi_dealloc_type_bool|qapi_dealloc_type_int64|qapi_dealloc_type_null|qapi_dealloc_type_number|qapi_dealloc_type_str|qapi_dealloc_type_uint64|qobject_input_check_list|qobject_input_check_struct|qobject_input_start_alternate|qobject_input_start_list|qobject_input_start_struct|qobject_input_type_any|qobject_input_type_bool|qobject_input_type_bool_keyval|qobject_input_type_int64|qobject_input_type_int64_keyval|qobject_input_type_null|qobject_input_type_number|qobject_input_type_number_keyval|qobject_input_type_size_keyval|qobject_input_type_str|qobject_input_type_str_keyval|qobject_input_type_uint64|qobject_input_type_uint64_keyval|qobject_output_start_list|qobject_output_start_struct|qobject_output_type_any|qobject_output_type_bool|qobject_output_type_int64|qobject_output_type_null|qobject_output_type_number|qobject_output_type_str|qobject_output_type_uint64|start_list|visit_check_list|visit_check_struct|visit_start_alternate|visit_start_list|visit_start_struct|visit_type_.*";
    expression list args;
    typedef Error;
    Error *err;
    @@
    -    fun(args, &err);
    -    if (err)
    +    if (!fun(args, &err))
         {
             ...
         }

A few line breaks tidied up manually.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>

qapi: Make visitor functions taking Error ** return bool, not void

See recent commit "error: Document Error API usage rules" for
rationale.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>

hmp: Eliminate a variable in hmp_migrate_set_parameter()

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>

block: Avoid error accumulation in bdrv_img_create()

When creating an image fails because the format doesn't support option
"backing_file" or "backing_fmt", bdrv_img_create() first has
qemu_opt_set() put a generic error into @local_err, then puts the real
error into @errp with error_setg(), and then propagates the former to
the latter, which throws away the generic error.  A bit complicated,
but works.

Now that qemu_opt_set() returns a useful value, we can simply ignore
the generic error instead.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>

qemu-option: Use returned bool to check for failure

The previous commit enables conversion of

    foo(..., &err);
    if (err) {
        ...
    }

to

    if (!foo(..., &err)) {
        ...
    }

for QemuOpts functions that now return true / false on success /
error.  Coccinelle script:

    @@
    identifier fun = {
        opts_do_parse, parse_option_bool, parse_option_number,
        parse_option_size, qemu_opt_parse, qemu_opt_rename, qemu_opt_set,
        qemu_opt_set_bool, qemu_opt_set_number, qemu_opts_absorb_qdict,
        qemu_opts_do_parse, qemu_opts_from_qdict_entry, qemu_opts_set,
        qemu_opts_validate
    };
    expression list args, args2;
    typedef Error;
    Error *err;
    @@
    -    fun(args, &err, args2);
    -    if (err)
    +    if (!fun(args, &err, args2))
         {
             ...
         }

A few line breaks tidied up manually.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>
[Conflict with commit 0b6786a9c1 "block/amend: refactor qcow2 amend
options" resolved by rerunning Coccinelle on master's version]

qemu-option: Make functions taking Error ** return bool, not void

See recent commit "error: Document Error API usage rules" for
rationale.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>

qemu-option: Replace opt_set() by cleaner opt_validate()

opt_set() frees its argument @value on failure.  Slightly unclean;
functions ideally do nothing on failure.

To tidy this up, move opt_create() from opt_set() into its callers,
along with the cleanup.  Rename opt_set() to opt_validate(), noting
its similarity to qemu_opts_validate().  Drop redundant parameter
@opts; use opt->opts instead.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>

qemu-option: Factor out helper opt_create()

There is just one use so far.  The next commit will add more.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>

qemu-option: Simplify around find_default_by_name()

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Message-Id: <20200707160613[email protected]>
Reviewed-by: Greg Kurz <[email protected]>

qemu-option: Factor out helper find_default_by_name()

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Reviewed-by: Greg Kurz <[email protected]>
Message-Id: <20200707160613[email protected]>

qemu-option: Make uses of find_desc_by_name() more similar

This is to make the next commit easier to review.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Reviewed-by: Greg Kurz <[email protected]>
Message-Id: <20200707160613[email protected]>

qemu-option: Check return value instead of @err where convenient

Convert uses like

    opts = qemu_opts_create(..., &err);
    if (err) {
        ...
    }

to

    opts = qemu_opts_create(..., errp);
    if (!opts) {
        ...
    }

Eliminate error_propagate() that are now unnecessary.  Delete @err
that are now unused.

Note that we can't drop parallels_open()'s error_propagate() here.  We
continue to execute it even in the converted case.  It's a no-op then:
local_err is null.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Reviewed-by: Greg Kurz <[email protected]>
Message-Id: <20200707160613[email protected]>

virtio-crypto-pci: Tidy up virtio_crypto_pci_realize()

virtio_crypto_pci_realize() continues after realization of its
"virtio-crypto-device" fails.  Only an object_property_set_link()
follows; looks harmless to me.  Tidy up anyway: return after failure,
just like virtio_rng_pci_realize() does.

Cc: "Gonglei (Arei)" <[email protected]>
Cc: Michael S. Tsirkin <[email protected]>
Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Reviewed-by: Gonglei < [email protected]>
Message-Id: <20200707160613[email protected]>

macio: Tidy up error handling in macio_newworld_realize()

macio_newworld_realize() effectively ignores ns->gpio realization
errors, leaking the Error object.  Fortunately, macio_gpio_realize()
can't actually fail.  Tidy up.

Cc: Mark Cave-Ayland <[email protected]>
Cc: David Gibson <[email protected]>
Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Acked-by: David Gibson <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Reviewed-by: Greg Kurz <[email protected]>
Message-Id: <20200707160613[email protected]>

qdev: Use returned bool to check for qdev_realize() etc. failure

Convert

    foo(..., &err);
    if (err) {
        ...
    }

to

    if (!foo(..., &err)) {
        ...
    }

for qdev_realize(), qdev_realize_and_unref(), qbus_realize() and their
wrappers isa_realize_and_unref(), pci_realize_and_unref(),
sysbus_realize(), sysbus_realize_and_unref(), usb_realize_and_unref().
Coccinelle script:

    @@
    identifier fun = {
        isa_realize_and_unref, pci_realize_and_unref, qbus_realize,
        qdev_realize, qdev_realize_and_unref, sysbus_realize,
        sysbus_realize_and_unref, usb_realize_and_unref
    };
    expression list args, args2;
    typedef Error;
    Error *err;
    @@
    -    fun(args, &err, args2);
    -    if (err)
    +    if (!fun(args, &err, args2))
         {
             ...
         }

Chokes on hw/arm/musicpal.c's lcd_refresh() with the unhelpful error
message "no position information".  Nothing to convert there; skipped.

Fails to convert hw/arm/armsse.c, because Coccinelle gets confused by
ARMSSE being used both as typedef and function-like macro there.
Converted manually.

A few line breaks tidied up manually.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Reviewed-by: Greg Kurz <[email protected]>
Message-Id: <20200707160613[email protected]>

error: Document Error API usage rules

This merely codifies existing practice, with one exception: the rule
advising against returning void, where existing practice is mixed.

When the Error API was created, we adopted the (unwritten) rule to
return void when the function returns no useful value on success,
unlike GError, which recommends to return true on success and false on
error then.

When a function returns a distinct error value, say false, a checked
call that passes the error up looks like

    if (!frobnicate(..., errp)) {
        handle the error...
    }

When it returns void, we need

    Error *err = NULL;

    frobnicate(..., &err);
    if (err) {
        handle the error...
        error_propagate(errp, err);
    }

Not only is this more verbose, it also creates an Error object even
when @errp is null, &error_abort or &error_fatal.

People got tired of the additional boilerplate, and started to ignore
the unwritten rule.  The result is confusion among developers about
the preferred usage.

Make the rule advising against returning void official by putting it
in writing.  This will hopefully reduce confusion.

Update the examples accordingly.

The remainder of this series will update a substantial amount of code
to honor the rule.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Reviewed-by: Greg Kurz <[email protected]>
Message-Id: <20200707160613[email protected]>
[Tweak prose as per advice from Eric]

error: Improve error.h's big comment

Add headlines to the big comment.

Explain examples for NULL, &error_abort and &error_fatal argument
better.

Tweak rationale for error_propagate_prepend().

Signed-off-by: Markus Armbruster <[email protected]>
Message-Id: <20200707160613[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Greg Kurz <[email protected]>

error: Fix examples in error.h's big comment

Mark a bad example more clearly.  Fix the error_propagate_prepend()
example.  Add a missing declaration and a second error pileup example.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Reviewed-by: Greg Kurz <[email protected]>
Message-Id: <20200707160613[email protected]>

Merge remote-tracking branch 'remotes/stefanha/tags/tracing-pull-request' into staging

Pull request

Fix for a LTTng Userspace Tracer header problem.

# gpg: Signature made Tue 07 Jul 2020 16:10:04 BST
# gpg:                using RSA key 8695A8BFD3F97CDAAC35775A9CA4ABB381AB73C8
# gpg: Good signature from "Stefan Hajnoczi <[email protected]>" [full]
# gpg:                 aka "Stefan Hajnoczi <[email protected]>" [full]
# Primary key fingerprint: 8695 A8BF D3F9 7CDA AC35  775A 9CA4 ABB3 81AB 73C8

* remotes/stefanha/tags/tracing-pull-request:
  tracetool: work around ust <sys/sdt.h> include conflict

Signed-off-by: Peter Maydell <[email protected]>

Merge remote-tracking branch 'remotes/philmd-gitlab/tags/fw_cfg-20200704' into staging

firmware (and crypto) patches

- add the tls-cipher-suites object,
- add the ability to QOM objects to produce data consumable
  by the fw_cfg device,
- let the tls-cipher-suites object implement the
  FW_CFG_DATA_GENERATOR interface.

This is required by EDK2 'HTTPS Boot' feature of OVMF to tell
the guest which TLS ciphers it can use.

CI jobs results:
  https://travis-ci.org/github/philmd/qemu/builds/704724619
  https://gitlab.com/philmd/qemu/-/pipelines/162938106
  https://cirrus-ci.com/build/4682977303068672

# gpg: Signature made Sat 04 Jul 2020 17:37:08 BST
# gpg:                using RSA key FAABE75E12917221DCFD6BB2E3E32C2CDEADC0DE
# gpg: Good signature from "Philippe Mathieu-Daudé (F4BUG) <[email protected]>" [full]
# Primary key fingerprint: FAAB E75E 1291 7221 DCFD  6BB2 E3E3 2C2C DEAD C0DE

* remotes/philmd-gitlab/tags/fw_cfg-20200704:
  crypto/tls-cipher-suites: Produce fw_cfg consumable blob
  softmmu/vl: Allow -fw_cfg 'gen_id' option to use the 'etc/' namespace
  softmmu/vl: Let -fw_cfg option take a 'gen_id' argument
  hw/nvram/fw_cfg: Add the FW_CFG_DATA_GENERATOR interface
  crypto: Add tls-cipher-suites object

Signed-off-by: Peter Maydell <[email protected]>

Merge remote-tracking branch 'remotes/kraxel/tags/modules-20200707-pull-request' into staging

qom: add support for qom objects in modules.
build some devices (qxl, virtio-gpu, ccid, usb-redir) as modules.
build braille chardev as module.

v2: more verbose comment for "build: fix device module builds" patch.

note: qemu doesn't rebuild objects on cflags changes (specifically
      -fPIC being added when code is switched from builtin to module).
      Workaround for resulting build errors: "make clean", rebuild.

# gpg: Signature made Tue 07 Jul 2020 14:42:16 BST
# gpg:                using RSA key 4CB6D8EED3E87138
# gpg: Good signature from "Gerd Hoffmann (work) <[email protected]>" [full]
# gpg:                 aka "Gerd Hoffmann <[email protected]>" [full]
# gpg:                 aka "Gerd Hoffmann (private) <[email protected]>" [full]
# Primary key fingerprint: A032 8CFF B93A 17A7 9901  FE7D 4CB6 D8EE D3E8 7138

* remotes/kraxel/tags/modules-20200707-pull-request:
  chardev: enable modules, use for braille
  vga: build virtio-gpu as module
  vga: build virtio-gpu only once
  vga: build qxl as module
  usb: build usb-redir as module
  ccid: build smartcard as module
  build: fix device module builds
  qdev: device module support
  object: qom module support
  module: qom module support

Signed-off-by: Peter Maydell <[email protected]>

Merge remote-tracking branch 'remotes/vivier2/tags/trivial-branch-for-5.1-pull-request' into staging

trivial branch patches 20200707

# gpg: Signature made Tue 07 Jul 2020 11:52:06 BST
# gpg:                using RSA key CD2F75DDC8E3A4DC2E4F5173F30C38BD3F2FBE3C
# gpg:                issuer "[email protected]"
# gpg: Good signature from "Laurent Vivier <[email protected]>" [full]
# gpg:                 aka "Laurent Vivier <[email protected]>" [full]
# gpg:                 aka "Laurent Vivier (Red Hat) <[email protected]>" [full]
# Primary key fingerprint: CD2F 75DD C8E3 A4DC 2E4F  5173 F30C 38BD 3F2F BE3C

* remotes/vivier2/tags/trivial-branch-for-5.1-pull-request:
  net/tap-solaris.c: Include qemu-common.h for TFR macro
  intel_iommu: "aw-bits" error message still refers to "x-aw-bits"
  util/qemu-option: Document the get_opt_value() function
  MAINTAINERS: Update Radoslaw Biernacki email address
  .mailmap: Update Alexander Graf email address
  trivial: Respect alphabetical order of .o files in Makefile.objs
  fix the prototype of muls64/mulu64

Signed-off-by: Peter Maydell <[email protected]>

Merge remote-tracking branch 'remotes/vivier/tags/m68k-next-pull-request' into staging

m68k pull-request 20200706

disable floatx80_invalid_encoding() for m68k
fix m68k_cpu_get_phys_page_debug()

# gpg: Signature made Mon 06 Jul 2020 21:05:33 BST
# gpg:                using RSA key CD2F75DDC8E3A4DC2E4F5173F30C38BD3F2FBE3C
# gpg:                issuer "[email protected]"
# gpg: Good signature from "Laurent Vivier <[email protected]>" [full]
# gpg:                 aka "Laurent Vivier <[email protected]>" [full]
# gpg:                 aka "Laurent Vivier (Red Hat) <[email protected]>" [full]
# Primary key fingerprint: CD2F 75DD C8E3 A4DC 2E4F  5173 F30C 38BD 3F2F BE3C

* remotes/vivier/tags/m68k-next-pull-request:
  softfloat,m68k: disable floatx80_invalid_encoding() for m68k
  target/m68k: consolidate physical translation offset into get_physical_address()
  target/m68k: fix physical address translation in m68k_cpu_get_phys_page_debug()

Signed-off-by: Peter Maydell <[email protected]>

Merge remote-tracking branch 'remotes/kraxel/tags/audio-20200706-pull-request' into staging

audio: deprecate -soundhw

# gpg: Signature made Mon 06 Jul 2020 20:29:07 BST
# gpg:                using RSA key 4CB6D8EED3E87138
# gpg: Good signature from "Gerd Hoffmann (work) <[email protected]>" [full]
# gpg:                 aka "Gerd Hoffmann <[email protected]>" [full]
# gpg:                 aka "Gerd Hoffmann (private) <[email protected]>" [full]
# Primary key fingerprint: A032 8CFF B93A 17A7 9901  FE7D 4CB6 D8EE D3E8 7138

* remotes/kraxel/tags/audio-20200706-pull-request:
  audio: set default value for pcspk.iobase property
  pcspk: update docs/system/target-i386-desc.rst.inc
  audio: add soundhw deprecation notice
  audio: deprecate -soundhw pcspk
  audio: create pcspk device early
  audio: rework pcspk_init()
  softmmu: initialize spice and audio earlier
  pc_basic_device_init: drop no_vmport arg
  pc_basic_device_init: drop has_pit arg
  pc_basic_device_init: pass PCMachineState
  audio: deprecate -soundhw hda
  audio: deprecate -soundhw sb16
  audio: deprecate -soundhw gus
  audio: deprecate -soundhw cs4231a
  audio: deprecate -soundhw adlib
  audio: deprecate -soundhw es1370
  audio: deprecate -soundhw ac97
  audio: add deprecated_register_soundhw
  stubs: add pci_create_simple
  stubs: add isa_create_simple

Signed-off-by: Peter Maydell <[email protected]>

Merge remote-tracking branch 'remotes/rth/tags/pull-tcg-20200706' into staging

Fix for ppc shifts
Fix for non-parallel atomic ops

# gpg: Signature made Mon 06 Jul 2020 19:49:08 BST
# gpg:                using RSA key 7A481E78868B4DB6A85A05C064DF38E8AF7E215F
# gpg:                issuer "[email protected]"
# gpg: Good signature from "Richard Henderson <[email protected]>" [full]
# Primary key fingerprint: 7A48 1E78 868B 4DB6 A85A  05C0 64DF 38E8 AF7E 215F

* remotes/rth/tags/pull-tcg-20200706:
  tcg: Fix do_nonatomic_op_* vs signed operations
  tcg/ppc: Sanitize immediate shifts

Signed-off-by: Peter Maydell <[email protected]>

Merge remote-tracking branch 'remotes/maxreitz/tags/pull-block-2020-07-06' into staging

Block patches for 5.1:
- LUKS keyslot amendment
  (+ patches to make the iotests pass on non-Linux systems, and to keep
     the tests passing for qcow v1, and to skip LUKS tests (including
     qcow2 LUKS) when the built qemu does not support it)
- Refactoring in the block layer: Drop the basically unnecessary
  unallocated_blocks_are_zero field from BlockDriverInfo
- Fix qcow2 preallocation when the image size is not a multiple of the
  cluster size
- Fix in block-copy code

# gpg: Signature made Mon 06 Jul 2020 11:02:53 BST
# gpg:                using RSA key 91BEB60A30DB3E8857D11829F407DB0061D5CF40
# gpg:                issuer "[email protected]"
# gpg: Good signature from "Max Reitz <[email protected]>" [full]
# Primary key fingerprint: 91BE B60A 30DB 3E88 57D1  1829 F407 DB00 61D5 CF40

* remotes/maxreitz/tags/pull-block-2020-07-06: (31 commits)
  qed: Simplify backing reads
  block: drop unallocated_blocks_are_zero
  block/vhdx: drop unallocated_blocks_are_zero
  block/file-posix: drop unallocated_blocks_are_zero
  block/iscsi: drop unallocated_blocks_are_zero
  block/crypto: drop unallocated_blocks_are_zero
  block/vpc: return ZERO block-status when appropriate
  block/vdi: return ZERO block-status when appropriate
  block: inline bdrv_unallocated_blocks_are_zero()
  qemu-img: convert: don't use unallocated_blocks_are_zero
  iotests: add tests for blockdev-amend
  block/qcow2: implement blockdev-amend
  block/crypto: implement blockdev-amend
  block/core: add generic infrastructure for x-blockdev-amend qmp command
  iotests: qemu-img tests for luks key management
  block/qcow2: extend qemu-img amend interface with crypto options
  block/crypto: implement the encryption key management
  block/crypto: rename two functions
  block/amend: refactor qcow2 amend options
  block/amend: separate amend and create options for qemu-img
  ...

Signed-off-by: Peter Maydell <[email protected]>

Merge remote-tracking branch 'remotes/mst/tags/for_upstream' into staging

virtio,acpi: features, fixes, cleanups.

vdpa support
virtio-mem support
a handy script for disassembling acpi tables
misc fixes and cleanups

Signed-off-by: Michael S. Tsirkin <[email protected]>
# gpg: Signature made Tue 07 Jul 2020 13:00:35 BST
# gpg:                using RSA key 5D09FD0871C8F85B94CA8A0D281F0DB8D28D5469
# gpg:                issuer "[email protected]"
# gpg: Good signature from "Michael S. Tsirkin <[email protected]>" [full]
# gpg:                 aka "Michael S. Tsirkin <[email protected]>" [full]
# Primary key fingerprint: 0270 606B 6F3C DF3D 0B17  0970 C350 3912 AFBE 8E67
#      Subkey fingerprint: 5D09 FD08 71C8 F85B 94CA  8A0D 281F 0DB8 D28D 5469

* remotes/mst/tags/for_upstream: (41 commits)
  vhost-vdpa: introduce vhost-vdpa net client
  vhost-vdpa: introduce vhost-vdpa backend
  vhost_net: introduce set_config & get_config
  vhost: implement vhost_force_iommu method
  vhost: introduce new VhostOps vhost_force_iommu
  vhost: implement vhost_vq_get_addr method
  vhost: introduce new VhostOps vhost_vq_get_addr
  vhost: implement vhost_dev_start method
  vhost: introduce new VhostOps vhost_dev_start
  vhost: check the existence of vhost_set_iotlb_callback
  virtio-pci: implement queue_enabled method
  virtio-bus: introduce queue_enabled method
  vhost_net: use the function qemu_get_peer
  net: introduce qemu_get_peer
  MAINTAINERS: add VT-d entry
  docs: vhost-user: add Virtio status protocol feature
  tests/acpi: remove stale allowed tables
  numa: Auto-enable NUMA when any memory devices are possible
  virtio-mem: Exclude unplugged memory during migration
  virtio-mem: Add trace events
  ...

Signed-off-by: Peter Maydell <[email protected]>
# Conflicts:
# hw/arm/virt.c
# hw/virtio/trace-events

tracetool: work around ust <sys/sdt.h> include conflict

Both the dtrace and ust backends may include <sys/sdt.h> but LTTng
Userspace Tracer 2.11 and later requires SDT_USE_VARIADIC to be defined
before including the header file.

This is a classic problem with C header files included from different
parts of a program. If the same header is included twice within the same
compilation unit then the first inclusion determines the macro
environment.

Work around this by defining SDT_USE_VARIADIC in the dtrace backend too.
It doesn't hurt and fixes a missing STAP_PROBEV() compiler error when
the ust backend is enabled together with the dtrace backend.

Signed-off-by: Stefan Hajnoczi <[email protected]>
Message-id: 20200625140757[email protected]
Signed-off-by: Stefan Hajnoczi <[email protected]>

chardev: enable modules, use for braille

Removes brlapi library dependency from core qemu.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200624131045[email protected]

vga: build virtio-gpu as module

Drops libvirglrenderer.so dependency from core qemu.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200624131045[email protected]

vga: build virtio-gpu only once

Signed-off-by: Gerd Hoffmann <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>
Message-id: 20200624131045[email protected]

vga: build qxl as module

First step in making spice support modular.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200624131045[email protected]

usb: build usb-redir as module

Drops libusbredirparser.so dependency from core qemu.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200624131045[email protected]

ccid: build smartcard as module

Drops libcacard.so dependency from core qemu.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200624131045[email protected]

build: fix device module builds

Slightly hackish workaround, works ok as long as we don't
have target-specific modules.  meson will obsolete this.

See comment in the patch for the --verbose description.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200624131045[email protected]

[ kraxel: updated comment from discussions ]

Signed-off-by: Gerd Hoffmann <[email protected]>

qdev: device module support

Hook module loading into the places where we
need it when building devices as modules.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200624131045[email protected]

object: qom module support

Little helper function to load modules on demand.  In most cases adding
module loading support for devices and other objects is just
s/object_class_by_name/module_object_class_by_name/ in the right spot.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200624131045[email protected]

module: qom module support

Add support for qom types provided by modules.  For starters use a
manually maintained list which maps qom type to module and prefix.

Two load functions are added:  One to load the module for a specific
type, and one to load all modules (needed for object/device lists as
printed by -- for example -- qemu -device help).

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200624131045[email protected]

vhost-vdpa: introduce vhost-vdpa net client

This patch set introduces a new net client type: vhost-vdpa.
vhost-vdpa net client will set up a vDPA device which is specified
by a "vhostdev" parameter.

Signed-off-by: Lingshan Zhu <[email protected]>
Signed-off-by: Tiwei Bie <[email protected]>
Signed-off-by: Cindy Lu <[email protected]>
Signed-off-by: Jason Wang <[email protected]>
Message-Id: <20200701145538[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Acked-by: Jason Wang <[email protected]>

vhost-vdpa: introduce vhost-vdpa backend

Currently we have 2 types of vhost backends in QEMU: vhost kernel and
vhost-user. The above patch provides a generic device for vDPA purpose,
this vDPA device exposes to user space a non-vendor-specific configuration
interface for setting up a vhost HW accelerator, this patch set introduces
a third vhost backend called vhost-vdpa based on the vDPA interface.

Vhost-vdpa usage:

qemu-system-x86_64 -cpu host -enable-kvm \
    ......
    -netdev type=vhost-vdpa,vhostdev=/dev/vhost-vdpa-id,id=vhost-vdpa0 \
    -device virtio-net-pci,netdev=vhost-vdpa0,page-per-vq=on \

Signed-off-by: Lingshan zhu <[email protected]>
Signed-off-by: Tiwei Bie <[email protected]>
Signed-off-by: Cindy Lu <[email protected]>
Signed-off-by: Jason Wang <[email protected]>
Message-Id: <20200701145538[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Acked-by: Jason Wang <[email protected]>

Merge remote-tracking branch 'remotes/huth-gitlab/tags/pull-request-2020-07-06' into staging

* Fuzzer fixes from Alexander
* Clean-up patches for qtests, configure and mcf5206
* Sparc64 sun4u acceptance test

# gpg: Signature made Mon 06 Jul 2020 08:34:14 BST
# gpg:                using RSA key 27B88847EEE0250118F3EAB92ED9D774FE702DB5
# gpg:                issuer "[email protected]"
# gpg: Good signature from "Thomas Huth <[email protected]>" [full]
# gpg:                 aka "Thomas Huth <[email protected]>" [full]
# gpg:                 aka "Thomas Huth <[email protected]>" [full]
# gpg:                 aka "Thomas Huth <[email protected]>" [unknown]
# Primary key fingerprint: 27B8 8847 EEE0 2501 18F3  EAB9 2ED9 D774 FE70 2DB5

* remotes/huth-gitlab/tags/pull-request-2020-07-06:
  tests/acceptance: Add a test for the sun4u sparc64 machine
  hw/m68k/mcf5206: Replace remaining hw_error()s by qemu_log_mask()
  configure / util: Auto-detect the availability of openpty()
  tests/qtest: Unify the test for the xenfv and xenpv machines
  fuzz: do not use POSIX shm for coverage bitmap
  fuzz: fix broken qtest check at rcu_disable_atfork

Signed-off-by: Peter Maydell <[email protected]>

net/tap-solaris.c: Include qemu-common.h for TFR macro

In commit a8d2532645cf5ce4 we cleaned up usage of the qemu-common.h header
so that it was always included from .c files and never from other .h files.
We missed adding it to net/tap-solaris.c (which previously was pulling it
in via tap-int.h), which broke building on Solaris hosts.

Fixes: a8d2532645cf5ce4
Reported-by: Michele Denber <[email protected]>
Signed-off-by: Peter Maydell <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Thomas Huth <[email protected]>
Tested-by: Michele Denber <[email protected]>
Message-Id: <20200704092317[email protected]>
Signed-off-by: Laurent Vivier <[email protected]>

intel_iommu: "aw-bits" error message still refers to "x-aw-bits"

Commit 4b49b586c4 ('intel_iommu: remove "x-" prefix for "aw-bits"')
removed the "x-" prefix but but didn't update the error message
accordingly.

Signed-off-by: Menno Lageman <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20200625155258.1452425[email protected]>
Signed-off-by: Laurent Vivier <[email protected]>

util/qemu-option: Document the get_opt_value() function

Coverity noticed commit 950c4e6c94 introduced a dereference before
null check in get_opt_value (CID1391003):

  In get_opt_value: All paths that lead to this null pointer
  comparison already dereference the pointer earlier (CWE-476)

We fixed this in commit 6e3ad3f0e31, but relaxed the check in commit
0c2f6e7ee99 because "No callers of get_opt_value() pass in a NULL
for the 'value' parameter".

Since this function is publicly exposed, it risks new users to do
the same error again. Avoid that documenting the 'value' argument
must not be NULL.

Signed-off-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Daniel P. Berrangé <[email protected]>
Message-Id: <20200629070858[email protected]>
Signed-off-by: Laurent Vivier <[email protected]>

MAINTAINERS: Update Radoslaw Biernacki email address

My Linaro account is no longer active and stop forwarding emails to me.
Changing it to my current employer domain.

Signed-off-by: Radoslaw Biernacki <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>
Acked-by: Leif Lindholm <[email protected]>
Message-Id: <20200512170704[email protected]>
Signed-off-by: Laurent Vivier <[email protected]>

.mailmap: Update Alexander Graf email address

Update Alexander Graf email address to avoid emails bouncing.

Suggested-by: Alexander Graf <[email protected]>
Signed-off-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Alexander Graf <[email protected]>
Message-Id: <20200702173818[email protected]>
Signed-off-by: Laurent Vivier <[email protected]>

trivial: Respect alphabetical order of .o files in Makefile.objs

The vmgenid.o is the only file that is not in alphabetical order.

Signed-off-by: Christophe de Dinechin <[email protected]>
Reviewed-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Message-Id: <20200629094934.2081180[email protected]>
Signed-off-by: Laurent Vivier <[email protected]>

softfloat,m68k: disable floatx80_invalid_encoding() for m68k

According to the comment, this definition of invalid encoding is given
by intel developer's manual, and doesn't comply with 680x0 FPU.

With m68k, the explicit integer bit can be zero in the case of:
 - zeros                (exp == 0, mantissa == 0)
 - denormalized numbers (exp == 0, mantissa != 0)
 - unnormalized numbers (exp != 0, exp < 0x7FFF)
 - infinities           (exp == 0x7FFF, mantissa == 0)
 - not-a-numbers        (exp == 0x7FFF, mantissa != 0)

For infinities and NaNs, the explicit integer bit can be either one or
zero.

The IEEE 754 standard does not define a zero integer bit. Such a number
is an unnormalized number. Hardware does not directly support
denormalized and unnormalized numbers, but implicitly supports them by
trapping them as unimplemented data types, allowing efficient conversion
in software.

See "M68000 FAMILY PROGRAMMER’S REFERENCE MANUAL",
    "1.6 FLOATING-POINT DATA TYPES"

We will implement in the m68k TCG emulator the FP_UNIMP exception to
trap into the kernel to normalize the number. In case of linux-user,
the number will be normalized by QEMU.

Signed-off-by: Laurent Vivier <[email protected]>
Reviewed-by: Alex Bennée <[email protected]>
Message-Id: <20200612140400.2130118[email protected]>

target/m68k: consolidate physical translation offset into get_physical_address()

Since all callers to get_physical_address() now apply the same page offset to
the translation result, move the logic into get_physical_address() itself to
avoid duplication.

Suggested-by: Philippe Mathieu-Daudé <[email protected]>
Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20200701201531[email protected]>
Signed-off-by: Laurent Vivier <[email protected]>

target/m68k: fix physical address translation in m68k_cpu_get_phys_page_debug()

The result of the get_physical_address() function should be combined with the
offset of the original page access before being returned. Otherwise the
m68k_cpu_get_phys_page_debug() function can round to the wrong page causing
incorrect lookups in gdbstub and various "Disassembler disagrees with
translator over instruction decoding" warnings to appear at translation time.

Fixes: 88b2fef6c3 ("target/m68k: add MC68040 MMU")
Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20200701201531[email protected]>
Signed-off-by: Laurent Vivier <[email protected]>

tcg: Fix do_nonatomic_op_* vs signed operations

The smin/smax/umin/umax operations require the operands to be
properly sign extended.  Do not drop the MO_SIGN bit from the
load, and additionally extend the val input.

Reviewed-by: LIU Zhiwei <[email protected]>
Reported-by: LIU Zhiwei <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>
Message-Id: <20200701165646.1901320[email protected]>

tcg/ppc: Sanitize immediate shifts

Sanitize shift constants so that shift operations with
large constants don't generate invalid instructions.

Signed-off-by: Catherine A. Frederick <[email protected]>
Message-Id: <20200607211100[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

fix the prototype of muls64/mulu64

The prototypes of muls64/mulu64 in host-utils.h should match the
definitions in host-utils.c

Signed-off-by: Lijun Pan <[email protected]>
Message-Id: <20200701234344[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Signed-off-by: Laurent Vivier <[email protected]>

audio: set default value for pcspk.iobase property

Allows dropping the explicit qdev_prop_set_uint32 call in pcspk_init.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200702132525[email protected]

pcspk: update docs/system/target-i386-desc.rst.inc

Add PC speaker with config hints.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200702132525[email protected]

audio: add soundhw deprecation notice

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200702132525[email protected]

audio: deprecate -soundhw pcspk

Add deprecation message to the audio init function.

Factor out audio initialization and call that from
both audio init and realize, so setting the audiodev
property is enough to properly initialize pcspk.

Add a property alias to the machine type to set the
audio device, so pcspk can be initialized using:
"-machine pcspk-audiodev=<name>"

Using "-global isa-pcspk.audiodev=<name>" works too but
is not recommended.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200702132525[email protected]

audio: create pcspk device early

Create the pcspk device early, so it exists at
machine type initialization time.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200702132525[email protected]

audio: rework pcspk_init()

Instead of creating and returning the pc speaker accept it as argument.
That allows to rework the initialization workflow in followup patches.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200702132525[email protected]

softmmu: initialize spice and audio earlier

audiodev must be initialized before machine_set_property
so the machine can have audiodev property aliases.

spice must initialize before audiodev because the default
audiodev is spice only in case spice is actually enabled.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200702132525[email protected]

pc_basic_device_init: drop no_vmport arg

Now that we pass pcms anyway, we don't need the no_vmport arg any more.
No functional change.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200702132525[email protected]

pc_basic_device_init: drop has_pit arg

Now that we pass pcms anyway, we don't need the has_pit arg any more.
No functional change.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200702132525[email protected]

pc_basic_device_init: pass PCMachineState

Need access to pcms for pcspk initialization.
Just preparation, no functional change.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200702132525[email protected]

audio: deprecate -soundhw hda

Add deprecation message to the audio init function.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200702132525[email protected]

audio: deprecate -soundhw sb16

Switch to deprecated_register_soundhw().
Remove the now obsolete init function.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200702132525[email protected]

audio: deprecate -soundhw gus

Switch to deprecated_register_soundhw().
Remove the now obsolete init function.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200702132525[email protected]

audio: deprecate -soundhw cs4231a

Switch to deprecated_register_soundhw().
Remove the now obsolete init function.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200702132525[email protected]

audio: deprecate -soundhw adlib

Switch to deprecated_register_soundhw().
Remove the now obsolete init function.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200702132525[email protected]

audio: deprecate -soundhw es1370

Switch to deprecated_register_soundhw().  Remove the now obsolete init
function.  Add an alias so both es1370 and ES1370 are working with
-device.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200702132525[email protected]

audio: deprecate -soundhw ac97

Switch to deprecated_register_soundhw().  Remove the now obsolete init
function.  Add an alias so both ac97 and AC97 are working with -device.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200702132525[email protected]

audio: add deprecated_register_soundhw

Add helper function for -soundhw deprecation.  It can replace the
simple init functions which just call {isa,pci}_create_simple()
with a hardcoded type.  It also prints a deprecation message.

Signed-off-by: Gerd Hoffmann <[email protected]>
Message-id: 20200702132525[email protected]