Re: [Qemu-devel] [PATCH 04/27] vl: convert -m to qemu_opts_parse()

[Markus Armbruster <armbru@redhat.com>]

Igor Mammedov <imammedo@redhat.com> writes:

> Along with conversion extend -m option to support following parameters:

Please split this into two patches: first conversion to QemuOpts, then
extension.

>   "mem" - startup memory amount
>   "slots" - total number of hotplug memory slots
>   "maxmem" - maximum possible memory
>
> "slots" and "maxmem" should go in pair and "maxmem" should be greater
> than "mem" for memory hotplug to be usable.
>
> v2:
>   make sure maxmem is not less than ram size
>
> Signed-off-by: Igor Mammedov <imammedo@redhat.com>
> ---
>  qemu-options.hx |    9 +++++-
>  vl.c            |   73 ++++++++++++++++++++++++++++++++++++++++++++++---------
>  2 files changed, 68 insertions(+), 14 deletions(-)
>
> diff --git a/qemu-options.hx b/qemu-options.hx
> index 8b94264..fe4559b 100644
> --- a/qemu-options.hx
> +++ b/qemu-options.hx
> @@ -210,8 +210,13 @@ use is discouraged as it may be removed from future versions.
>  ETEXI
>  
>  DEF("m", HAS_ARG, QEMU_OPTION_m,
> -    "-m megs         set virtual RAM size to megs MB [default="
> -    stringify(DEFAULT_RAM_SIZE) "]\n", QEMU_ARCH_ALL)
> +    "-m [mem=]megs[,slots=n,maxmem=size]\n"
> +    "                set virtual RAM size to megs MB [default="
> +    stringify(DEFAULT_RAM_SIZE) "]\n"
> +    "                mem=start-up memory amount\n"
> +    "                slots=maximum number of hotplug slots\n"
> +    "                maxmem=maximum total amount of memory\n",
> +    QEMU_ARCH_ALL)

Help text is confusing, because it discusses megs twice.  Fits right in,
as output of -help is generally confusing (to put it politely).

What about something like this:

-m [mem=]megs[,slots=n,maxmem=size]
                configure guest RAM
                mem: initial amount of guest memory (default: XXX)
                slots: number of hotplug slots (default: none)
                maxmem: maximum amount of guest memory (default: mem)

>  STEXI
>  @item -m @var{megs}
>  @findex -m
> diff --git a/vl.c b/vl.c
> index f28674f..5974f0f 100644
> --- a/vl.c
> +++ b/vl.c
> @@ -529,6 +529,28 @@ static QemuOptsList qemu_msg_opts = {
>      },
>  };
>  
> +static QemuOptsList qemu_mem_opts = {
> +    .name = "memory-opts",
> +    .implied_opt_name = "mem",
> +    .head = QTAILQ_HEAD_INITIALIZER(qemu_mem_opts.head),
> +    .merge_lists = true,

Yes, because multiple -m need to accumulate.

> +    .desc = {
> +        {
> +            .name = "mem",
> +            .type = QEMU_OPT_SIZE,
> +        },
> +        {
> +            .name = "slots",
> +            .type = QEMU_OPT_NUMBER,
> +        },
> +        {
> +            .name = "maxmem",
> +            .type = QEMU_OPT_SIZE,
> +        },
> +        { /* end of list */ }
> +    },
> +};
> +
>  /**
>   * Get machine options
>   *
> @@ -2816,6 +2838,14 @@ static int object_create(QemuOpts *opts, void *opaque)
>      return 0;
>  }
>  
> +static void qemu_init_default_mem_opts(uint64_t size)
> +{
> +    QemuOpts *opts = qemu_opts_create_nofail(&qemu_mem_opts);
> +    qemu_opt_set_number(opts, "mem", size);
> +    qemu_opt_set_number(opts, "maxmem", size);
> +    qemu_opt_set_number(opts, "slots", 0);
> +}
> +

We usually don't put defaults in QemuOpts.  Instead, the code querying
QemuOpts detects and handles absence of the parameter.  Can be as simple
as qemu_opt_get_size(opts, "mem", DEFAULT_RAM_SIZE * 1024 * 1024).

See also below.

>  int main(int argc, char **argv, char **envp)
>  {
>      int i;
> @@ -2887,6 +2917,7 @@ int main(int argc, char **argv, char **envp)
>      qemu_add_opts(&qemu_tpmdev_opts);
>      qemu_add_opts(&qemu_realtime_opts);
>      qemu_add_opts(&qemu_msg_opts);
> +    qemu_add_opts(&qemu_mem_opts);
>  
>      runstate_init();
>  
> @@ -2901,7 +2932,8 @@ int main(int argc, char **argv, char **envp)
>      module_call_init(MODULE_INIT_MACHINE);
>      machine = find_default_machine();
>      cpu_model = NULL;
> -    ram_size = 0;
> +    ram_size = DEFAULT_RAM_SIZE * 1024 * 1024;
> +    qemu_init_default_mem_opts(ram_size);
>      snapshot = 0;
>      cyls = heads = secs = 0;
>      translation = BIOS_ATA_TRANSLATION_AUTO;
> @@ -3178,21 +3210,43 @@ int main(int argc, char **argv, char **envp)
>                  exit(0);
>                  break;
>              case QEMU_OPTION_m: {
> -                int64_t value;
>                  uint64_t sz;
> -                char *end;
> +                const char *end;
> +                char *s;
>  
> -                value = strtosz(optarg, &end);
> -                if (value < 0 || *end) {
> -                    fprintf(stderr, "qemu: invalid ram size: %s\n", optarg);
> +                opts = qemu_opts_parse(qemu_find_opts("memory-opts"),
> +                                       optarg, 1);
> +                if (!opts) {
>                      exit(1);
>                  }
> -                sz = QEMU_ALIGN_UP((uint64_t)value, 8192);
> +
> +                /* fixup legacy sugffix-less format */

/* fix up legacy suffix-less format */

The problem here is that OPT_SIZE treats values without a size suffix as
bytes, but we need to default to MiB for backward compatibility.

> +                end = qemu_opt_get(opts, "mem");
> +                if (g_ascii_isdigit(end[strlen(end) - 1])) {
> +                    s = g_strconcat(end, "M", NULL);
> +                    qemu_opt_set(opts, "mem", s);
> +                    g_free(s);
> +                }

Ugly.

Why is the variable called 'end'?

qemu_opt_set() appends to the list of options.  The un-fixed-up option
remains in the list.  qemu_opt_unset() could fix that, but it asserts
opts_accepts_any() for unclear reasons.  git-blame points to Kevin.

Have you considered qemu_opt_set_number()?

If this "need a default suffix" pattern exists elsewhere, we should
consider extending QemuOptDesc to cover it.

> +
> +                sz = QEMU_ALIGN_UP(qemu_opt_get_size(opts, "mem", ram_size),
> +                                   8192);
> +                /* compatibility behaviour for case "-m 0" */
> +                if (sz == 0) {
> +                    sz = DEFAULT_RAM_SIZE * 1024 * 1024;
> +                }
> +

Yes, this is needed.  Our command line is bonkers.

>                  ram_size = sz;
>                  if (ram_size != sz) {
>                      fprintf(stderr, "qemu: ram size too large\n");
>                      exit(1);
>                  }
> +                /* store aligned value for future use */
> +                qemu_opt_set_number(opts, "mem", ram_size);

Here, you use qemu_opt_set_number().

Again, this appends to the list, and leaves the non-aligned value in.

> +
> +                sz = qemu_opt_get_size(opts, "maxmem", ram_size);
> +                if (sz < ram_size) {
> +                    qemu_opt_set_number(opts, "maxmem", ram_size);
> +                }
>                  break;
>              }

Looks like you want to fix up something like "-m 1024", so that maxmem
stays equal to mem.  I'm afraid you also "fix" user errors like "-m
mem=1024M,maxmem=512M".

If you refrain from putting defaults into opts, you can distinguish the
cases "user didn't specify maxmem, so assume mem", and "user specified
maxmem, so check it's >= mem".

>  #ifdef CONFIG_TPM
> @@ -4029,11 +4083,6 @@ int main(int argc, char **argv, char **envp)
>          exit(1);
>      }
>  
> -    /* init the memory */
> -    if (ram_size == 0) {
> -        ram_size = DEFAULT_RAM_SIZE * 1024 * 1024;
> -    }
> -
>      if (qemu_opts_foreach(qemu_find_opts("device"), device_help_func, NULL, 0)
>          != 0) {
>          exit(0);