[PATCH 1/1] acpi: cannot have RSDT above 4 GiB

[Simon Glass]

Hi Heinrich,

On Sun, 12 Nov 2023 at 13:16, Heinrich Schuchardt
<heinrich.schuchardt at canonical.com> wrote:
>
>
>
> Simon Glass <sjg at chromium.org> schrieb am So., 12. Nov. 2023, 21:03:
>>
>> Hi Heinrich,
>>
>> On Sat, 11 Nov 2023 at 07:28, Heinrich Schuchardt
>> <heinrich.schuchardt at canonical.com> wrote:
>> >
>> > The field RsdtAddress has only 32 bit. The RSDT table cannot be located
>> > beyond 4 GiB.
>> >
>> > Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt at canonical.com>
>> > ---
>> >  lib/acpi/base.c | 26 +++++++++++++++++++-------
>> >  1 file changed, 19 insertions(+), 7 deletions(-)
>>
>> Reviewed-by: Simon Glass <sjg at chromium.org>
>>
>> nits / question below
>>
>> >
>> > diff --git a/lib/acpi/base.c b/lib/acpi/base.c
>> > index 2057bd2bef..128a76ad39 100644
>> > --- a/lib/acpi/base.c
>> > +++ b/lib/acpi/base.c
>> > @@ -21,10 +21,17 @@ void acpi_write_rsdp(struct acpi_rsdp *rsdp, struct acpi_rsdt *rsdt,
>> >         memcpy(rsdp->signature, RSDP_SIG, 8);
>> >         memcpy(rsdp->oem_id, OEM_ID, 6);
>> >
>> > -       rsdp->length = sizeof(struct acpi_rsdp);
>> > -       rsdp->rsdt_address = map_to_sysmem(rsdt);
>> > +       if (rsdt)
>> > +               rsdp->rsdt_address = map_to_sysmem(rsdt);
>> > +       else
>> > +               rsdp->rsdt_address = 0;
>>
>> There is a memset() at the top so this line (and the one below) should
>> not be needed.
>>
>> > +
>> > +       if (xsdt)
>> > +               rsdp->xsdt_address = map_to_sysmem(xsdt);
>> > +       else
>> > +               rsdp->xsdt_address = 0;
>> >
>> > -       rsdp->xsdt_address = map_to_sysmem(xsdt);
>> > +       rsdp->length = sizeof(struct acpi_rsdp);
>> >         rsdp->revision = ACPI_RSDP_REV_ACPI_2_0;
>> >
>> >         /* Calculate checksums */
>> > @@ -68,11 +75,15 @@ static void acpi_write_xsdt(struct acpi_xsdt *xsdt)
>> >  static int acpi_write_base(struct acpi_ctx *ctx,
>> >                            const struct acpi_writer *entry)
>> >  {
>> > -       /* We need at least an RSDP and an RSDT Table */
>> > +       /* We need at least an RSDP and an XSDT Table */
>> >         ctx->rsdp = ctx->current;
>> >         acpi_inc_align(ctx, sizeof(struct acpi_rsdp));
>> > -       ctx->rsdt = ctx->current;
>> > -       acpi_inc_align(ctx, sizeof(struct acpi_rsdt));
>> > +       if (map_to_sysmem(ctx->current) < UINT_MAX - 0x10000) {
>>
>> Won't that do something different on 64-bit machines? It seems that
>> you want to check against SZ_4G ?
>
>
> For gcc int is 32bit on 64bit systems, so is UINT_MAX. Using SZ_4G could make this code clearer.

Oh I didn't know that...I just assumed that uint would be 64-bit on a
64-machine machine, since it is the natural reg size! Yes, that is
very very confusing. I vaguely recall some compiler setting for that
and that Linux did it a particular way.

>
> Best regards
>
> Heinrich
>
>
>>
>> > +               ctx->rsdt = ctx->current;
>> > +               acpi_inc_align(ctx, sizeof(struct acpi_rsdt));
>> > +       } else {
>> > +               ctx->rsdt = 0;
>> > +       }
>> >         ctx->xsdt = ctx->current;
>> >         acpi_inc_align(ctx, sizeof(struct acpi_xsdt));
>> >
>> > @@ -80,7 +91,8 @@ static int acpi_write_base(struct acpi_ctx *ctx,
>> >         memset(ctx->base, '\0', ctx->current - ctx->base);
>> >
>> >         acpi_write_rsdp(ctx->rsdp, ctx->rsdt, ctx->xsdt);
>> > -       acpi_write_rsdt(ctx->rsdt);
>> > +       if (ctx->rsdt)
>> > +               acpi_write_rsdt(ctx->rsdt);
>> >         acpi_write_xsdt(ctx->xsdt);
>> >
>> >         return 0;
>> > --
>> > 2.40.1
>> >
>>
>> Regards,
>> Simon