Re: [PATCH V5] acpi/prmt: find block with specific type

[Ard Biesheuvel]

On Thu, 3 Oct 2024 at 05:45, Koba Ko <kobak@xxxxxxxxxx> wrote:
>
>
> On 10/3/24 05:11, Ard Biesheuvel wrote:
> > External email: Use caution opening links or attachments
> >
> >
> > On Wed, 2 Oct 2024 at 20:06, Rafael J. Wysocki <rafael@xxxxxxxxxx> wrote:
> >> On Wed, Sep 11, 2024 at 5:55 PM KobaK <kobak@xxxxxxxxxx> wrote:
> >>> PRMT needs to find the correct type of block to
> >>> translate the PA-VA mapping for EFI runtime services.
> >>>
> >>> The issue arises because the PRMT is finding a block of
> >>> type EFI_CONVENTIONAL_MEMORY, which is not appropriate for
> >>> runtime services as described in Section 2.2.2 (Runtime
> >>> Services) of the UEFI Specification [1]. Since the PRM handler is
> >>> a type of runtime service, this causes an exception
> >>> when the PRM handler is called.
> >>>
> >>>      [Firmware Bug]: Unable to handle paging request in EFI runtime service
> >>>      WARNING: CPU: 22 PID: 4330 at drivers/firmware/efi/runtime-wrappers.c:341
> >>>          __efi_queue_work+0x11c/0x170
> >>>      Call trace:
> >>>        __efi_queue_work+0x11c/0x170
> >>>        efi_call_acpi_prm_handler+0x68/0xd0
> >>>        acpi_platformrt_space_handler+0x198/0x258
> >>>        acpi_ev_address_space_dispatch+0x144/0x388
> >>>        acpi_ex_access_region+0x9c/0x118
> >>>        acpi_ex_write_serial_bus+0xc4/0x218
> >>>        acpi_ex_write_data_to_field+0x168/0x218
> >>>        acpi_ex_store_object_to_node+0x1a8/0x258
> >>>        acpi_ex_store+0xec/0x330
> >>>        acpi_ex_opcode_1A_1T_1R+0x15c/0x618
> >>>        acpi_ds_exec_end_op+0x274/0x548
> >>>        acpi_ps_parse_loop+0x10c/0x6b8
> >>>        acpi_ps_parse_aml+0x140/0x3b0
> >>>        acpi_ps_execute_method+0x12c/0x2a0
> >>>        acpi_ns_evaluate+0x210/0x310
> >>>        acpi_evaluate_object+0x178/0x358
> >>>        acpi_proc_write+0x1a8/0x8a0 [acpi_call]
> >>>        proc_reg_write+0xcc/0x150
> >>>        vfs_write+0xd8/0x380
> >>>        ksys_write+0x70/0x120
> >>>        __arm64_sys_write+0x24/0x48
> >>>        invoke_syscall.constprop.0+0x80/0xf8
> >>>        do_el0_svc+0x50/0x110
> >>>        el0_svc+0x48/0x1d0
> >>>        el0t_64_sync_handler+0x15c/0x178
> >>>        el0t_64_sync+0x1a8/0x1b0
> >>>
> >>> Find a block with specific type to fix this.
> >>> prmt find a block with EFI_RUNTIME_SERVICES_DATA for prm handler and
> >>> find a block with EFI_RUNTIME_SERVICES_CODE for prm context.
> >>> If no suitable block is found, a warning message will be prompted
> >>> but the procedue continues to manage the next prm handler.
> >>> However, if the prm handler is actullay called without proper allocation,
> >>> it would result in a failure during error handling.
> >>>
> >>> By using the correct memory types for runtime services,
> >>> Ensure that the PRM handler and the context are
> >>> properly mapped in the virtual address space during runtime,
> >>> preventing the paging request error.
> >>>
> >>> [1] https://uefi.org/sites/default/files/resources/UEFI_Spec_2_10_Aug29.pdf
> >> I need input from EFI people on this, so can you please resend the
> >> patch with a CC to linux-efi@xxxxxxxxxxxxxxx?
> >>
> >>> Fixes: cefc7ca46235 ("ACPI: PRM: implement OperationRegion handler for the PlatformRtMechanism subtype")
> >>> Signed-off-by: KobaK <kobak@xxxxxxxxxx>
> > Please use your full name.
> Hi Ardb,
> Sure, will update.
> >
> >>> Reviewed-by: Matthew R. Ochs <mochs@xxxxxxxxxx>
> >>> ---
> >>> V2:
> >>> 1. format the changelog and add more about error handling.
> >>> 2. replace goto
> >>> V3: Warn if parts of handler are missed during va-pa translating.
> >>> V4: Fix the 0day
> >>> V5: Fix typo and pr_warn warning
> >>> ---
> >>>   drivers/acpi/prmt.c | 49 +++++++++++++++++++++++++++++++--------------
> >>>   1 file changed, 34 insertions(+), 15 deletions(-)
> >>>
> >>> diff --git a/drivers/acpi/prmt.c b/drivers/acpi/prmt.c
> >>> index c78453c74ef5..cd4a7f5491d6 100644
> >>> --- a/drivers/acpi/prmt.c
> >>> +++ b/drivers/acpi/prmt.c
> >>> @@ -72,15 +72,17 @@ struct prm_module_info {
> >>>          struct prm_handler_info handlers[] __counted_by(handler_count);
> >>>   };
> >>>
> >>> -static u64 efi_pa_va_lookup(u64 pa)
> >>> +static u64 efi_pa_va_lookup(u64 pa, u32 type)
> >>>   {
> >>>          efi_memory_desc_t *md;
> >>>          u64 pa_offset = pa & ~PAGE_MASK;
> >>>          u64 page = pa & PAGE_MASK;
> >>>
> >>>          for_each_efi_memory_desc(md) {
> >>> -               if (md->phys_addr < pa && pa < md->phys_addr + PAGE_SIZE * md->num_pages)
> >>> +               if ((md->type == type) &&
> >>> +                       (md->phys_addr < pa && pa < md->phys_addr + PAGE_SIZE * md->num_pages)) {
> >>>                          return pa_offset + md->virt_addr + page - md->phys_addr;
> >>> +               }
> >>>          }
> >>>
> >>>          return 0;
> >>> @@ -148,9 +150,18 @@ acpi_parse_prmt(union acpi_subtable_headers *header, const unsigned long end)
> >>>                  th = &tm->handlers[cur_handler];
> >>>
> >>>                  guid_copy(&th->guid, (guid_t *)handler_info->handler_guid);
> >>> -               th->handler_addr = (void *)efi_pa_va_lookup(handler_info->handler_address);
> >>> -               th->static_data_buffer_addr = efi_pa_va_lookup(handler_info->static_data_buffer_address);
> >>> -               th->acpi_param_buffer_addr = efi_pa_va_lookup(handler_info->acpi_param_buffer_address);
> >>> +               th->handler_addr =
> >>> +                       (void *)efi_pa_va_lookup(handler_info->handler_address, EFI_RUNTIME_SERVICES_CODE);
> > Wouldn't it make more sense to test the EFI_MEMORY_RUNTIME attribute
> > rather than expecting/assuming a certain memory type in each case?
> > That attribute is precisely what controls whether or not a region has
> > been remapped into the firmware's page tables.
> Please see the below
> >
> >>> +               th->static_data_buffer_addr =
> >>> +                       efi_pa_va_lookup(handler_info->static_data_buffer_address, EFI_RUNTIME_SERVICES_DATA);
> >>> +               th->acpi_param_buffer_addr =
> >>> +                       efi_pa_va_lookup(handler_info->acpi_param_buffer_address, EFI_RUNTIME_SERVICES_DATA);
> >>> +
> >>> +               if (!th->handler_addr || !th->static_data_buffer_addr || !th->acpi_param_buffer_addr)
> >>> +                       pr_warn(
> >>> +                               "Idx: %llu, Parts of handler(GUID: %pUL) are missed, handler_addr %p, data_addr %p, param_addr %p",
> > Please improve this diagnostic: 'are missed' is not very helpful.
>
> Are these good for you
>

I /think/ it looks ok but please resend it as a proper patch - the
whitespace got mangled and it is difficult to read.

> ```
>
> -static u64 efi_pa_va_lookup(u64 pa, u32 type)
> +static u64 efi_pa_va_lookup(u64 pa)
>   {
>          efi_memory_desc_t *md;
>          u64 pa_offset = pa & ~PAGE_MASK;
>          u64 page = pa & PAGE_MASK;
>
>          for_each_efi_memory_desc(md) {
> -               if ((md->type == type) &&
> +               if ((md->attribute & EFI_MEMORY_RUNTIME) &&
>                          (md->phys_addr < pa && pa < md->phys_addr +
> PAGE_SIZE * md->num_pages)) {
>                          return pa_offset + md->virt_addr + page -
> md->phys_addr;
>                  }
> @@ -150,18 +150,20 @@ acpi_parse_prmt(union acpi_subtable_headers
> *header, const unsigned long end)
>                  th = &tm->handlers[cur_handler];
>
>                  guid_copy(&th->guid, (guid_t *)handler_info->handler_guid);
> -               th->handler_addr =
> -                       (void
> *)efi_pa_va_lookup(handler_info->handler_address,
> EFI_RUNTIME_SERVICES_CODE);
> -               th->static_data_buffer_addr =
> - efi_pa_va_lookup(handler_info->static_data_buffer_address,
> EFI_RUNTIME_SERVICES_DATA);
> -               th->acpi_param_buffer_addr =
> - efi_pa_va_lookup(handler_info->acpi_param_buffer_address,
> EFI_RUNTIME_SERVICES_DATA);
> -
> -               if (!th->handler_addr || !th->static_data_buffer_addr ||
> !th->acpi_param_buffer_addr)
> -                       pr_warn(
> -                               "Idx: %llu, Parts of handler(GUID: %pUL)
> are missed, handler_addr %p, data_addr %p, param_addr %p",
> -                               cur_handler, &th->guid, th->handler_addr,
> -                               (void *)th->static_data_buffer_addr,
> (void *)th->acpi_param_buffer_addr);
> +               th->handler_addr = (void
> *)efi_pa_va_lookup(handler_info->handler_address);
> +               if (!th->handler_addr)
> +                       pr_warn( "Idx: %llu, failed to find VA for
> handler_addr(GUID: %pUL, PA: %p)",
> +                               cur_handler, &th->guid, th->handler_addr);
> +
> +               th->static_data_buffer_addr =
> efi_pa_va_lookup(handler_info->static_data_buffer_address);
> +               if (!th->static_data_buffer_addr)
> +                       pr_warn( "Idx: %llu, failed to find VA for
> data_addr(PA: %p)",
> +                               cur_handler, &th->guid, (void
> *)th->static_data_buffer_addr);
> +
> +               th->acpi_param_buffer_addr =
> efi_pa_va_lookup(handler_info->acpi_param_buffer_address);
> +               if (!th->acpi_param_buffer_addr)
> +                       pr_warn( "Idx: %llu, failed to find VA for
> param_addr(PA: %p)",
> +                               cur_handler, &th->guid, (void
> *)th->acpi_param_buffer_addr);
>
> ```
>
...
> >>> @@ -274,8 +293,10 @@ static acpi_status acpi_platformrt_space_handler(u32 function,
> >>>          case PRM_CMD_START_TRANSACTION:
> >>>
> >>>                  module = find_prm_module(&buffer->handler_guid);
> >>> -               if (!module)
> >>> -                       goto invalid_guid;
> >>> +               if (!module) {
> >>> +                       buffer->prm_status = PRM_HANDLER_GUID_NOT_FOUND;
> >>> +                       return AE_OK;
> >>> +               }
> > What is the reason for this change, and the ones down below?
> As per Rui's comment, goto can be replaced with return.
> So I modified them with return and PRM_HANDLER_GUID_NOT_FOUND.

I don't think this change is necessary, but it should be a separate
patch at the very least.