Re: [Qemu-devel] [PATCH v2 06/47] acpi: add acpi_name() & acpi_name_decl() term

[Igor Mammedov]

On Fri, 23 Jan 2015 15:42:30 +0200
"Michael S. Tsirkin" <address@hidden> wrote:

> On Fri, Jan 23, 2015 at 02:32:45PM +0100, Igor Mammedov wrote:
> > On Fri, 23 Jan 2015 10:59:48 +0200
> > "Michael S. Tsirkin" <address@hidden> wrote:
> > 
> > > On Thu, Jan 22, 2015 at 02:49:50PM +0000, Igor Mammedov wrote:
> > > > Signed-off-by: Igor Mammedov <address@hidden>
> > > > ---
> > > >  hw/acpi/acpi-build-utils.c         | 24 ++++++++++++++++++++++++
> > > >  include/hw/acpi/acpi-build-utils.h |  3 +++
> > > >  2 files changed, 27 insertions(+)
> > > > 
> > > > diff --git a/hw/acpi/acpi-build-utils.c b/hw/acpi/acpi-build-utils.c
> > > > index 40a1769..1bda2ec 100644
> > > > --- a/hw/acpi/acpi-build-utils.c
> > > > +++ b/hw/acpi/acpi-build-utils.c
> > > > @@ -314,6 +314,30 @@ static AcpiAml aml_allocate_internal(uint8_t op, 
> > > > AcpiBlockFlags flags)
> > > >      return var;
> > > >  }
> > > >  
> > > > +/*
> > > > + * help to construct NameString, which return AcpiAml object
> > > > + * for using with other aml_append or other acpi_* terms
> > > 
> > > Here and elsewhere: I can't parse this header text.
> > > I'm guessing you just mean "construct NameString",
> > > and that's it?
> > yes
> > 
> > > 
> > > Also, most other places use build_append_namestring -
> > > so when should acpi_name be used instead?
> > > This should be made clear here in the comment.
> > acpi_name() is a replacement/wrapper around build_append_namestring()
> > which returns AcpiAml object. build_append_namestring() is a nonpublic
> > lowlevel helper that deals with GArray,
> > while acpi_name() follows semantic of AML API.
> > 
> > > 
> > > > + */
> > > > +AcpiAml GCC_FMT_ATTR(1, 2) acpi_name(const char *name_format, ...)
> > > > +{
> > > 
> > > This isn't really a name. It just appends a string.  So rename this
> > > acpi_string and then the below one adding a name can be named acpi_name?
> > acpi_string is introduced in 27/47, which is a prefixed string
> > as described in spec.
> > 
> > > Also, in many places one must use only one nameseg.
> > Where is it exactly?
> > Perhaps we could build in acpi_name() a check if we know in
> > what context enforce it. Better to have single/uniform API
> > for names than a several which is confusing.
> 
> I agree here.
> 
> > > I think a separate api that actually validates
> > > that it's one segment is better than silently failing.
> > > Do we ever use it for more than 1 segment?
> > Yes we use names with more than one segment.
> 
> Interesting. where exactly?
For example:

build_append_namestring(method, "^S%.02X.PCNT"

> 
> > > If not, maybe the right thing to do is
> > > to use build_append_nameseg and call this one acpi_nameseg.
> > acpi_name() is used only for passing name as arguments to methods,
> > in spec there isn't a limitation to only one segment when it comes
> > to names, in ASL part of it. namesegment however only AML construct
> > which helps to build name, I prefer not expose lowlevel AML
> > unless we have to.
> 
> OK, I agree.
> 
> > > 
> > > 
> > > > +    va_list ap;
> > > > +    AcpiAml var = aml_allocate_internal(0, NON_BLOCK);
> > > 
> > > 0 hard coded? What does it mean?
> > 1st arg for NON_BLOCK context doesn't mean anything/ignored.
> > alternatively I can make aml_allocate_nonblock() wrapper
> > around generic allocator.
> 
> 0 isn't a valid opcode either, it can really be anything.
yes, it could.
So I leave it as it's
or is there any other way that you'd prefer here?

> 
> 
> 
> > > Same elsewhere.
> > > 
> > > > +    va_start(ap, name_format);
> > > > +    build_append_namestringv(var.buf, name_format, ap);
> > > > +    va_end(ap);
> > > > +    return var;
> > > > +
> > > > +/* ACPI 5.0: 20.2.5.1 Namespace Modifier Objects Encoding: DefName */
> > > 
> > > Let's quote the earliest spec which documents each object:
> > > one year from now 5.0 will not be the latest.
> > > Applies here and elsewhere.
> > > In most places this will be 1.0b.
> > > Where the construct is newer, this will automatically
> > > document which guests support it.
> > I'll try to do it.
> > 
> > > 
> > > > +AcpiAml acpi_name_decl(const char *name, AcpiAml val)
> > > > +{
> > > > +    AcpiAml var = aml_allocate_internal(0, NON_BLOCK);
> > > > +    build_append_byte(var.buf, 0x08);
> > > 
> > > Pls add comment documenting what 0x08 is here.
> > sure
> > 
> > > 
> > > > +    build_append_namestring(var.buf, "%s", name);
> > > > +    aml_append(&var, val);
> > > > +    return var;
> > > > +}
> > > > +
> > > >  /* ACPI 5.0: 20.2.5.3 Type 1 Opcodes Encoding: DefIfElse */
> > > >  AcpiAml acpi_if(AcpiAml predicate)
> > > >  {
> > > > diff --git a/include/hw/acpi/acpi-build-utils.h 
> > > > b/include/hw/acpi/acpi-build-utils.h
> > > > index 177f9ed..868cfa5 100644
> > > > --- a/include/hw/acpi/acpi-build-utils.h
> > > > +++ b/include/hw/acpi/acpi-build-utils.h
> > > > @@ -21,6 +21,9 @@ typedef struct AcpiAml {
> > > >  
> > > >  void aml_append(AcpiAml *parent_ctx, AcpiAml child);
> > > >  
> > > > +/* non block ASL object primitives */
> > > 
> > > what does it mean that it's a "non block primitive"?
> > > I didn't find this concept in the spec.
> > As for a question what is NON_BLOCK, it's for simple inline ASL
> > construct that doesn't have to be packaged in special way
> > examles:
> >   Store(A,B)
> >   Name(FOO, VAL)
> >   IO(...)
> > while there are different block elements differing in how
> > they are created see 1/47 aml_append():
> > 
> > ResourceTemplate {
> >  /* block of other ASL items */
> > }
> > 
> > Package() {
> >  /* block of other ASL items */
> > }
> > 
> > if ... else ...
> > 
> > Scope() {
> >  /* block of other ASL items */
> > }
> > 
> > and so on.
> 
> "special way" is kind of vague.
> Maybe add a comment explaining when it's used.
> Is it when length isn't used as a prefix?
> AML_NO_PREFIX?
There are terms that have 'prefix' in definition but are still
inline elements, like:
DWordConst := DWordPrefix DWordData
String := StringPrefix AsciiCharList NullChar

using prefix would be confusing.

Having(/not having) prefix doesn't make objects into block one,
that can embed other objects. And although length-less is the closest
definition for them but it also doesn't match all objects, for
example:
NamedField := NameSeg PkgLength
uses PkgLength but still isn't a block element that includes
other objects.

That's why I've used NON_BLOCK as a neutral name but still
describing what's going on.

> 
> 
> > > 
> > > > +AcpiAml GCC_FMT_ATTR(1, 2) acpi_name(const char *name_format, ...);
> > > > +AcpiAml acpi_name_decl(const char *name, AcpiAml val);
> > > >  /* Block ASL object primitives */
> > > >  AcpiAml acpi_if(AcpiAml predicate);
> > > >  AcpiAml acpi_method(const char *name, int arg_count);
> > > > -- 
> > > > 1.8.3.1