Re: [PATCH 3/3] ACPI: Evaluate _CRS while creating device node objects

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Wednesday, November 14, 2012 10:18:46 AM Rafael J. Wysocki wrote:
> On Wednesday, November 14, 2012 02:23:51 AM Moore, Robert wrote:
> > Rafael,
> > 
> > I sounds like with a few changes, we can enhance this mechanism to
> > be more useful to you and others. Some comments below. I need to look
> > at the code in question a bit more, but I see no insurmountable issues.
> 
> Great, thanks!
>  
> 
> > > -----Original Message-----
> > > From: Rafael J. Wysocki [mailto:rjw@sisk.pl]
> > > Sent: Tuesday, November 13, 2012 2:57 PM
> > > To: Moore, Robert
> > > Cc: Mika Westerberg; mathias.nyman@linux.intel.com; linux-
> > > acpi@vger.kernel.org; linux-kernel@vger.kernel.org; lenb@kernel.org;
> > > Wysocki, Rafael J; broonie@opensource.wolfsonmicro.com;
> > > grant.likely@secretlab.ca; linus.walleij@linaro.org; khali@linux-fr.org;
> > > Bjorn Helgaas; Zheng, Lv
> > > Subject: Re: [PATCH 3/3] ACPI: Evaluate _CRS while creating device node
> > > objects
> > > 
> > > On Tuesday, November 13, 2012 10:06:03 PM Moore, Robert wrote:
> > > > I may not quite understand what you are asking for, but I will try.
> > > > It seems like we already have much of what you want/need, so maybe I'm
> > > > missing something.
> > > 
> > > I think all of the necessary pieces are there.
> > > 
> > > > > So what I would like to have, in general terms, is something like
> > > > > acpi_walk_resources() split into three parts:
> > > > >
> > > > >  (1) One that processes the _CRS output and creates a list of
> > > > >      struct acpi_resource objects for us to play with.  I suppose
> > > > >      it's OK if that's just a buffer filled with resource objects,
> > > > >      but a linked list might be more convenient.
> > > > >
> > > >
> > > > This sounds like AcpiGetCurrentResources. It executes _CRS and formats
> > > > the data into acpi_resource objects.
> > > 
> > > Yes, it does.  However, it is not completely clear to me if/how the caller
> > > is supposed to prepare the buffer object pointed to by the second arg.
> > > 
> > > If the buffer is initialized by AcpiGetCurrentResources, then that's what
> > > I need for (1).
> > 
> > 
> > It looks to me that at least AcpiGetCurrentResources does not actually ever
> > allocate a buffer for the resource template, it expects the caller to
> > eventually provide one of at least the size of the returned resource template.
> > 
> > This is really quite a bit out-of-date as far as the memory allocation model.
> > It should also support the option to just allocate the buffer of the appropriate
> > size before returning it to the caller.
> 
> Yes, that would be really useful.
> 
> Ideally, I'd like to be able to pass a pointer to an uninitialized buffer
> structure to it (or to a wrapper around it) and get a buffer full of
> struct acpi_resource objects (if _CRS returns any) back from it. :-)

Of course, I can add such a wrapper in the Linux-specific code just fine.


> > > > >  (2) One that allows us to access (read/write) resources in the
> > > > >      list returned by (1).  We don't need to open code walking
> > > > >      the list and I probably wouldn't event want to do that.  What
> > > > >      we need is to be able to walk the same list for a number of
> > > > >      times and possibly to modify values in the resource objects
> > > > >      if there are conflicts.
> > > >
> > > > This sounds like AcpiWalkResources. I suppose a possible issue is that
> > > > currently, AcpiWalkResources actually invokes the _CRS, _PRS, or _AEI
> > > > method on behalf of the caller.
> > > 
> > > Yes, that exactly is the problem.
> > > 
> > > > It might make more sense to allow the caller to pass in the resource
> > > > buffer returned from a call to _CRS, etc.
> > > 
> > > Yes! :-)
> > 
> > 
> > I'll take a closer look at this tomorrow.
> 
> Cool, thanks!
> 
> 
> > > > >  (3) One allowing us to free the list returned by (1) if not needed
> > > > >      any more.
> > > > >
> > > >
> > > > AcpiGetCurrentResources: Currently, everything is returned in a single
> > > > buffer to minimize the number of allocations. A buffer you can free
> > > > when you are done with it.
> > > 
> > > I suppose I should use ACPI_FREE(buffer.pointer) for that, but isn't it
> > > for the ACPICA's internal use only?
> > > 
> > > Besides, I would prefer to be able to pass just "buffer" for freeing,
> > > without having to touch its internals.  No big deal, but it would be
> > > nicer. :-)
> > 
> > 
> > The ACPI_BUFFER type is in fact a public type that is meant to return both the
> > buffer and the (actual) length. You will find many instances of
> > ACPI_FREE(buffer.pointer) within existing linux code, since it also used for
> > objects returned by control method execution/object evaluation.
> 
> Well, I suppose I only wanted to say that acpi_free_buffer(buffer) would look
> a bit more straightforward than ACPI_FREE(buffer.pointer). :-)

And I can define acpi_free_buffer() in the Linux-specific code too.

Thanks,
Rafael


-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.