Re: [PATCH 2/3] qom: move user_creatable_add_opts logic to vl.c and QAPIfy it

[Kevin Wolf]

Am 11.03.2021 um 18:24 hat Paolo Bonzini geschrieben:
> Emulators are currently using OptsVisitor (via user_creatable_add_opts)
> to parse the -object command line option.  This has one extra feature,
> compared to keyval, which is automatic conversion of integers to lists
> as well as support for lists as repeated options:
> 
>   -object 
> memory-backend-ram,id=pc.ram,size=1048576000,host-nodes=0,policy=bind
> 
> So we cannot replace OptsVisitor with keyval right now.  Still, this
> patch moves the user_creatable_add_opts logic to vl.c since it is
> not needed anywhere else, and makes it go through user_creatable_add_qapi.
> 
> In order to minimize code changes, the predicate still takes a string.
> This can be changed later to use the ObjectType QAPI enum directly.
> 
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

> diff --git a/softmmu/vl.c b/softmmu/vl.c
> index ff488ea3e7..b245e912e5 100644
> --- a/softmmu/vl.c
> +++ b/softmmu/vl.c
> @@ -117,6 +117,7 @@
>  #include "qapi/qapi-commands-block-core.h"
>  #include "qapi/qapi-commands-migration.h"
>  #include "qapi/qapi-commands-misc.h"
> +#include "qapi/qapi-visit-qom.h"
>  #include "qapi/qapi-commands-ui.h"
>  #include "qapi/qmp/qerror.h"
>  #include "sysemu/iothread.h"
> @@ -132,10 +133,16 @@ typedef struct BlockdevOptionsQueueEntry {
>  
>  typedef QSIMPLEQ_HEAD(, BlockdevOptionsQueueEntry) BlockdevOptionsQueue;
>  
> +typedef struct ObjectOption {
> +    ObjectOptions *opts;
> +    QTAILQ_ENTRY(ObjectOption) next;
> +} ObjectOption;
> +
>  static const char *cpu_option;
>  static const char *mem_path;
>  static const char *incoming;
>  static const char *loadvm;
> +static QTAILQ_HEAD(, ObjectOption) object_opts = 
> QTAILQ_HEAD_INITIALIZER(object_opts);
>  static ram_addr_t maxram_size;
>  static uint64_t ram_slots;
>  static int display_remote;
> @@ -1684,6 +1691,48 @@ static int machine_set_property(void *opaque,
>      return object_parse_property_opt(opaque, name, value, "type", errp);
>  }
>  
> +static void object_option_foreach_add(bool (*type_opt_predicate)(const char 
> *), Error **errp)
> +{
> +    ObjectOption *opt, *next;
> +
> +    QTAILQ_FOREACH_SAFE(opt, &object_opts, next, next) {
> +        const char *type = ObjectType_str(opt->opts->qom_type);
> +        if (type_opt_predicate(type)) {
> +            if (!user_creatable_add_qapi(opt->opts, errp)) {
> +                return;

Technically, this leaks things, though all callers pass &error_fatal
anyway, so this branch is dead code.

Should we just drop the errp parameter and explicitly use &error_fatal
here?

> +            }
> +            qapi_free_ObjectOptions(opt->opts);
> +            QTAILQ_REMOVE(&object_opts, opt, next);
> +        }
> +    }
> +}
> +
> +static void object_option_parse(const char *optarg)
> +{
> +    ObjectOption *opt;
> +    QemuOpts *opts;
> +    const char *type;
> +    Visitor *v;
> +
> +    opts = qemu_opts_parse_noisily(qemu_find_opts("object"),
> +                                   optarg, true);
> +    if (!opts) {
> +        exit(1);
> +    }
> +
> +    type = qemu_opt_get(opts, "qom-type");
> +    if (user_creatable_print_help(type, opts)) {

type needs a NULL check.

Before this patch:
$ build/qemu-system-x86_64 -object id=foo
qemu-system-x86_64: -object id=foo: Parameter 'qom-type' is missing

After the patch:
$ build/qemu-system-x86_64 -object id=foo
Segmentation fault (core dumped)

> +        exit(0);
> +    }
> +
> +    opt = g_new0(ObjectOption, 1);
> +    v = opts_visitor_new(opts);
> +    visit_type_ObjectOptions(v, NULL, &opt->opts, &error_fatal);
> +    visit_free(v);
> +
> +    QTAILQ_INSERT_TAIL(&object_opts, opt, next);
> +}

Kevin