ACPI: usage of sandbox virtual addresses

[Heinrich Schuchardt]

Am 29. Dezember 2023 15:31:24 MEZ schrieb Simon Glass <sjg at chromium.org>:
>Hi Heinrich,
>
>On Fri, Dec 29, 2023 at 11:14 AM Heinrich Schuchardt <xypron.glpk at gmx.de> wrote:
>>
>> On 12/29/23 09:26, Simon Glass wrote:
>> > Hi Heinrich,
>> >
>> > On Tue, Dec 26, 2023 at 10:03 AM Heinrich Schuchardt <xypron.glpk at gmx.de> wrote:
>> >>
>> >> On 12/26/23 10:50, Simon Glass wrote:
>> >>> Hi Heinrich,
>> >>>
>> >>> On Tue, Dec 26, 2023 at 8:56 AM Heinrich Schuchardt <xypron.glpk at gmx.de> wrote:
>> >>>>
>> >>>> Hello Simon,
>> >>>>
>> >>>> currently we use sandbox virtual addresses in all ACPI tables. This
>> >>>> means that an application started by the U-Boot sandbox consuming these
>> >>>> tables will crash due to accessing invalid addresses.
>> >>>>
>> >>>> Shouldn't the ACPI tables be migrated to use valid pointers?
>> >>>
>> >>> The confusion arises due to the difference between addresses and
>> >>> pointers. If we store addresses in the ACPI tables, then things should
>> >>> work.
>> >>>
>> >>> My approach has been to avoid using casts, but instead use
>> >>> map_sysmem() and mem_to_sysmem().
>> >>>
>> >>> Which particular piece of code is wrong in this case?
>> >>
>> >> Tables DSDT, XSDT, RSDP, FADT in the sandbox contain sandbox virtual
>> >> addresses instead of pointers to real host memory.
>> >>
>> >> All code referring to these tables should be changed.433:                            *gnvs = map_to_sysmem(ctx->current);
>> >
>> > I'm still not quite clear on this...can you point to functions or
>> > lines of code? When I look at acpi_add_table() it does use memmap, but
>> > perhaps some parts are wrong?
>> >
>> > Regards,
>> > Simon
>>
>> Here are some examples where the wrong values are set. We must get rid
>> of all map_to_sysmem() calls where writing ACPI tables.
>>
>> lib/acpi/acpi_table.c:
>> 170         rsdt->entry[i] = map_to_sysmem(table);
>> 188         xsdt->entry[i] = map_to_sysmem(table);
>>
>> lib/acpi/base.c:
>> 27:             rsdp->rsdt_address = map_to_sysmem(rsdt);
>> 30:             rsdp->xsdt_address = map_to_sysmem(xsdt);
>>
>> arch/x86/cpu/baytrail/acpi.c:
>> 81:     fadt->x_firmware_ctrl = map_to_sysmem(ctx->facs);
>> 82:     fadt->x_dsdt = map_to_sysmem(ctx->dsdt);
>>
>> arch/x86/cpu/quark/acpi.c:
>> 76:     fadt->x_firmware_ctrl = map_to_sysmem(ctx->facs);
>> 77:     fadt->x_dsdt = map_to_sysmem(ctx->dsdt);
>>
>> arch/x86/cpu/tangier/acpi.c:
>> 47:     fadt->x_firmware_ctrl = map_to_sysmem(ctx->facs);
>> 48:     fadt->x_dsdt = map_to_sysmem(ctx->dsdt);
>>
>> arch/x86/lib/acpi_table.c:
>> 200:    tcpa->lasa = map_to_sysmem(log);
>> 271:    tpm2->lasa = map_to_sysmem(lasa);
>> 433:                            *gnvs = map_to_sysmem(ctx->current);
>> 575:    fadt->x_firmware_ctrl = map_to_sysmem(facs);
>> 576:    fadt->x_dsdt = map_to_sysmem(dsdt);
>
>OK, I see. But at least within sandbox, the address is what we want to
>store, not the pointer. Are you worried about what an EFI app will do
>in that case, when we run an app from sandbox? If so, then yes it is
>definitely a problem.
>
>Regards,
>Simon

The sandbox is an environment to run EFI binaries inside an OS. So the ACPI tables must contain pointers.

The acpi command should display sandbox virtual addresses. All conversions should be moved to the command.

Best regards

Heinrich