Re: [PATCH v6 2/5] softmmu/vl: Let -fw_cfg option take a 'blob_id' argument

["Philippe Mathieu-Daudé" <philmd@redhat.com>]

On 5/20/20 12:34 AM, Laszlo Ersek wrote:
> On 05/19/20 20:20, Philippe Mathieu-Daudé wrote:
>> The 'blob_id' argument refers to a QOM object able to produce
>> data consumable by the fw_cfg device. The producer object must
>> implement the FW_CFG_DATA_GENERATOR interface.
> 
> OK, this answers my OBJECT_CHECK() question under patch #1 (in the
> negative -- an assert would be wrong).
> 
>>
>> Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
>> ---
>>  softmmu/vl.c | 17 +++++++++++++----
>>  1 file changed, 13 insertions(+), 4 deletions(-)
>>
>> diff --git a/softmmu/vl.c b/softmmu/vl.c
>> index ae5451bc23..f76c53ad2e 100644
>> --- a/softmmu/vl.c
>> +++ b/softmmu/vl.c
>> @@ -489,6 +489,10 @@ static QemuOptsList qemu_fw_cfg_opts = {
>>              .name = "string",
>>              .type = QEMU_OPT_STRING,
>>              .help = "Sets content of the blob to be inserted from a string",
>> +        }, {
>> +            .name = "blob_id",
>> +            .type = QEMU_OPT_STRING,
>> +            .help = "Sets id of the object generating fw_cfg blob to be used",
>>          },
>>          { /* end of list */ }
>>      },
>> @@ -2020,7 +2024,7 @@ static int parse_fw_cfg(void *opaque, QemuOpts *opts, Error **errp)
>>  {
>>      gchar *buf;
>>      size_t size;
>> -    const char *name, *file, *str;
>> +    const char *name, *file, *str, *blob_id;
>>      FWCfgState *fw_cfg = (FWCfgState *) opaque;
>>
>>      if (fw_cfg == NULL) {
>> @@ -2030,14 +2034,17 @@ static int parse_fw_cfg(void *opaque, QemuOpts *opts, Error **errp)
>>      name = qemu_opt_get(opts, "name");
>>      file = qemu_opt_get(opts, "file");
>>      str = qemu_opt_get(opts, "string");
>> +    blob_id = qemu_opt_get(opts, "blob_id");
>>
>>      /* we need name and either a file or the content string */
> 
> (1) Please update this comment. If the option is given, we need the
> name, and exactly one of: file, content string, blob_id.
> 
>> -    if (!(nonempty_str(name) && (nonempty_str(file) || nonempty_str(str)))) {
>> +    if (!(nonempty_str(name)
>> +          && (nonempty_str(file) || nonempty_str(str) || nonempty_str(blob_id)))
>> +         ) {
>>          error_setg(errp, "invalid argument(s)");
>>          return -1;
>>      }
> 
> (2) Coding style: does QEMU keep operators on the left or on the right
> when breaking subconditions to new lines? (I vaguely recall "to the
> right", but I could be wrong... Well, "hw/nvram/fw_cfg.c" has at least 7
> examples of the operator being on the right.)

You are right, I have been confused with a recommendation from Eric
Blake, but it was about shell script, not C.

> 
>> -    if (nonempty_str(file) && nonempty_str(str)) {
>> -        error_setg(errp, "file and string are mutually exclusive");
>> +    if (nonempty_str(file) && nonempty_str(str) && nonempty_str(blob_id)) {
>> +        error_setg(errp, "file, string and blob_id are mutually exclusive");
>>          return -1;
>>      }
> 
> (3) I believe this catches only when all three of name/string/blob_id
> are given. But we should continue catching "two given".
> 
> How about reworking both "if"s, *and* the comment at (1) at the same
> time, into:
> 
>     if (!nonempty_str(name) ||
>         nonempty_str(file) + nonempty_str(str) + nonempty_str(blob_id) != 1) {
>         error_setg(errp, "name, plus exactly one of file, string and blob_id, "
>                    "are needed");
>         return -1;
>     }
> 
> (Regarding the addition, nonempty_str() returns a "bool", which is a
> macro to _Bool, which is promoted to "int" or "unsigned int".)
> 
>>      if (strlen(name) > FW_CFG_MAX_FILE_PATH - 1) {
>> @@ -2052,6 +2059,8 @@ static int parse_fw_cfg(void *opaque, QemuOpts *opts, Error **errp)
>>      if (nonempty_str(str)) {
>>          size = strlen(str); /* NUL terminator NOT included in fw_cfg blob */
>>          buf = g_memdup(str, size);
>> +    } else if (nonempty_str(blob_id)) {
>> +        return fw_cfg_add_from_generator(fw_cfg, name, blob_id, errp);
>>      } else {
>>          GError *err = NULL;
>>          if (!g_file_get_contents(file, &buf, &size, &err)) {
>>
> 
> (4) The "-fw_cfg" command line option is documented in both the qemu(1)
> manual, and the "docs/specs/fw_cfg.txt" file.
> 
> I think we may have to update those. In particular I mean *where* the
> option is documented (in both texts).

Oh I forgot this part, thanks.

> 
> In the manual, "-fw_cfg" is currently under "Debug/Expert options", but
> that will no longer apply (I think?) after this series.

Well I'm not sure, the intent is the same, targeting mostly libvirt as
management interface; other uses are for "experts".

> 
> Similarly, in "docs/specs/fw_cfg.txt", the section is called "Externally
> Provided Items" -- but that might not be strictly true any more either.
> 
> Maybe leave the current "-fw_cfg" mentions in peace, and document
> "-fw_cfg blob_id=..." separately (in different docs sections)? The
> "fw_cfg generators" concept could deserve dedicated sections.
> 
> Sorry that I can't make a good concrete suggestion. :(

Thanks for the detailed review!

> 
> Thanks,
> Laszlo
>