Re: [Qemu-devel] [PATCH v3 01/52] acpi: introduce AML composer aml_append()

[Igor Mammedov]

On Tue, 17 Feb 2015 17:26:16 +0100
"Michael S. Tsirkin" <address@hidden> wrote:

> On Mon, Feb 09, 2015 at 10:53:23AM +0000, Igor Mammedov wrote:
> > Adds for dynamic AML creation, which will be used
> > for piecing ASL/AML primitives together and hiding
> > from user/caller details about how nested context
> > should be closed/packed leaving less space for
> > mistakes and necessity to know how AML should be
> > encoded, allowing user to concentrate on ASL
> > representation instead.
> > 
> > For example it will allow to create AML like this:
> > 
> > init_aml_allocator();
> > ...
> > Aml *scope = aml_scope("PCI0")
> > Aml *dev = aml_device("PM")
> >     aml_append(dev, aml_name_decl("_ADR", aml_int(addr)))
> > aml_append(scope, dev);
> > ...
> > free_aml_allocator();
> > 
> > Signed-off-by: Igor Mammedov <address@hidden>
> > ---
> >  hw/acpi/aml-build.c         | 91 
> > +++++++++++++++++++++++++++++++++++++++++++++
> >  hw/i386/acpi-build.c        |  1 -
> >  include/hw/acpi/aml-build.h | 61 ++++++++++++++++++++++++++++++
> >  3 files changed, 152 insertions(+), 1 deletion(-)
> > 
> > diff --git a/hw/acpi/aml-build.c b/hw/acpi/aml-build.c
> > index bcb288e..096f347 100644
> > --- a/hw/acpi/aml-build.c
> > +++ b/hw/acpi/aml-build.c
> > @@ -25,6 +25,8 @@
> >  #include <stdbool.h>
> >  #include <string.h>
> >  #include "hw/acpi/aml-build.h"
> > +#include "qemu/bswap.h"
> > +#include "hw/acpi/bios-linker-loader.h"
> >  
> >  GArray *build_alloc_array(void)
> >  {
> > @@ -257,3 +259,92 @@ void build_append_int(GArray *table, uint32_t value)
> >          build_append_value(table, value, 4);
> >      }
> >  }
> > +
> > +static GPtrArray *alloc_list;
> > +
> > +static Aml *aml_alloc(void)
> > +{
> > +    Aml *var = g_new0(typeof(*var), 1);
> > +
> > +    g_ptr_array_add(alloc_list, var);
> > +    var->block_flags = AML_HELPER;
> > +    var->buf = build_alloc_array();
> > +    return var;
> > +}
> > +
> > +static void aml_free(gpointer data)
> > +{
> > +    Aml *var = data;
> > +    build_free_array(var->buf);
> > +}
> > +
> > +Aml *init_aml_allocator(GArray *linker)
> > +{
> > +    Aml *var;
> > +
> > +    assert(!alloc_list);
> > +    alloc_list = g_ptr_array_new_with_free_func(aml_free);
> > +    var = aml_alloc();
> > +    var->linker = linker;
> > +    return var;
> > +}
> > +
> > +void free_aml_allocator(void)
> > +{
> > +    g_ptr_array_free(alloc_list, true);
> > +    alloc_list = 0;
> > +}
> > +
> > +static void build_buffer(GArray *array, uint8_t op)
> > +{
> > +    GArray *data = build_alloc_array();
> > +
> > +    build_append_int(data, array->len);
> > +    g_array_prepend_vals(array, data->data, data->len);
> > +    build_free_array(data);
> > +    build_package(array, op);
> > +}
> > +
> > +void aml_append(Aml *parent_ctx, Aml *child)
> > +{
> > +    switch (child->block_flags) {
> > +    case AML_NON_BLOCK:
> > +        build_append_byte(parent_ctx->buf, child->op);
> > +        break;
> > +    case AML_EXT_PACKAGE:
> > +        build_extop_package(child->buf, child->op);
> > +        break;
> > +    case AML_PACKAGE:
> > +        build_package(child->buf, child->op);
> > +        break;
> > +    case AML_RES_TEMPLATE:
> > +        build_append_byte(child->buf, 0x79); /* EndTag */
> > +        /*
> > +         * checksum operations is treated as succeeded if checksum
> > +         * field is zero. [ACPI Spec 5.0, 6.4.2.9 End Tag]
> > +         */
> > +        build_append_byte(child->buf, 0);
> > +        /* fall through, to pack resources in buffer */
> > +    case AML_BUFFER:
> > +        build_buffer(child->buf, child->op);
> > +        break;
> > +    case AML_DEF_BLOCK: {
> 
> This one is inelegant.
> It's not a definition block, and the patching and
it's definition block according to spec.

> manual offset calculation are scary.
it's the same thing as every user of build_header does every time,
so it's less scary compared to build_header and done only in one
place so user won't ever have to care about offsets.

> 
> > +        uint8_t *start = (uint8_t *)parent_ctx->buf->data +
> > +                         parent_ctx->buf->len;
> > +        uint32_t le32_len = cpu_to_le32(child->buf->len);
> > +
> > +        /* create linker entry for the DefinitionBlock */
> > +        bios_linker_loader_add_checksum(parent_ctx->linker,
> > +            ACPI_BUILD_TABLE_FILE,
> 
> Hard-coding file name here is very ugly.
> All this does not belong in this file, this
> is fw cfg interface.
It's but, it's the only one file and so far we don't have
plans to extend it and it does the same thing as build_header()


> 
> 
> > +            parent_ctx->buf->data,
> > +            start, child->buf->len, start + 9 /* checksum offset */);
> > +
> > +        /* set DefinitionBlock length at TableLength offset*/
> > +        memcpy(child->buf->data + 4, &le32_len, sizeof le32_len);
> 
> 
> This make no sense.
> In AML, there are no objects which are higher level
> than the table.
> 
> So for now, please just leave this alone:
> drop linker completely, keep table_data
> a simple GArray, not AML, and reuse build_header
> to fill in headers.
> 
> This way this file just deals with AML and ACPI spec.
Using build_header() is problematic for 2 reasons:
  1. it has to be done backwards,
     i.e. one has to reserve space for header first,
     than add content and then call build_header
     to patch header with current length and table definition data

  2. when using build_header() user has to calculate offsets and lengths
     manually every time.

While using aml_def_block() process is quite straightforward
and follows the declarative flow as it is in ASL,
which less confusing and error-prone, like:

ssdt = aml_def_block(...);
// add content to ssdt
aml_append(tables_blob, ssdt);

so hiding linker from user in this case makes usage simpler
and more harder to screw up, which is the whole point of
introducing this API.

I have a further plans to make other tables use
aml_def_block and get rid of acpi_add_table() which is also
confusing and simplify RSDT building doing it transparently
for user.

That way both ARM and i386 targets would reuse the same code
/API instead of re-implementing it in every target.
User will just get complete tables blob from API and
put it in respective fw_cfg file.
I also wish that we have had only one file for ACPI tables
(including RSDP) but ship already went away, maybe we can do
it for ARM though.

PS:
My secret goal is to bury linker and use it only when one
have to do so :), seriously.

> 
> 
> > +        break;
> > +    }
> > +    default:
> 
> This should assert.
sure

> 
> > +        break;
> > +    }
> > +    build_append_array(parent_ctx->buf, child->buf);
> > +}
> > diff --git a/hw/i386/acpi-build.c b/hw/i386/acpi-build.c
> > index 6d84f38..237080f 100644
> > --- a/hw/i386/acpi-build.c
> > +++ b/hw/i386/acpi-build.c
> > @@ -258,7 +258,6 @@ static void acpi_get_pci_info(PcPciInfo *info)
> >  #define ACPI_BUILD_APPNAME6 "BOCHS "
> >  #define ACPI_BUILD_APPNAME4 "BXPC"
> >  
> > -#define ACPI_BUILD_TABLE_FILE "etc/acpi/tables"
> >  #define ACPI_BUILD_RSDP_FILE "etc/acpi/rsdp"
> >  #define ACPI_BUILD_TPMLOG_FILE "etc/tpm/log"
> >  
> > diff --git a/include/hw/acpi/aml-build.h b/include/hw/acpi/aml-build.h
> > index 199f003..4033796 100644
> > --- a/include/hw/acpi/aml-build.h
> > +++ b/include/hw/acpi/aml-build.h
> > @@ -5,6 +5,67 @@
> >  #include <glib.h>
> >  #include "qemu/compiler.h"
> >  
> > +#define ACPI_BUILD_TABLE_FILE "etc/acpi/tables"
> > +
> > +typedef enum {
> > +    AML_HELPER = 0,
> > +    AML_NON_BLOCK,
> > +    AML_PACKAGE,
> > +    AML_EXT_PACKAGE,
> > +    AML_BUFFER,
> > +    AML_RES_TEMPLATE,
> > +    AML_DEF_BLOCK,
> > +} AmlBlockFlags;
> > +
> > +struct Aml {
> > +    GArray *buf;
> > +
> > +    /*< private >*/
> > +    uint8_t op;
> > +    AmlBlockFlags block_flags;
> > +    GArray *linker;
> > +};
> > +typedef struct Aml Aml;
> > +
> > +/**
> > + * init_aml_allocator:
> > + * @linker: linker that used by API for registering ACPI tables
> > + *          with linker firmware interfac
> > + *
> > + * Called for initializing API allocator which allow to use
> > + * AML API.
> > + * Returns: toplevel container which accumulates all other
> > + * ACPI tables.
> > + */
> > +Aml *init_aml_allocator(GArray *linker);
> > +
> > +/**
> > + * free_aml_allocator:
> > + *
> > + * Releases all elements used by AML API, frees associated memory
> > + * and invalidates AML allocator. After this call @init_aml_allocator
> > + * should be called again if AML API is to be used again.
> > + */
> > +void free_aml_allocator(void);
> > +
> > +/**
> > + * aml_append:
> > + * @parent_ctx: context to which @child element is added
> > + * @child: element that is copied into @parent_ctx context
> > + *
> > + * Joins Aml elements together and helps to construct AML tables
> > + * Examle of usage:
> > + *   Aml *table = aml_def_block("SSDT", ...);
> > + *   Aml *sb = aml_scope("\_SB");
> > + *   Aml *dev = aml_device("PCI0");
> > + *
> > + *   aml_append(dev, aml_name_decl("HID", aml_eisaid("PNP0A03")));
> > + *   aml_append(sb, dev);
> > + *   aml_append(table, sb);
> > + */
> > +void aml_append(Aml *parent_ctx, Aml *child);
> > +
> > +/* other helpers */
> >  GArray *build_alloc_array(void);
> >  void build_free_array(GArray *array);
> >  void build_prepend_byte(GArray *array, uint8_t val);
> > -- 
> > 1.8.3.1
>