On 18.10.18 09:57, AKASHI Takahiro wrote:
> On Wed, Oct 17, 2018 at 10:40:26AM +0200, Alexander Graf wrote:
>>
>>
>> On 17.10.18 09:32, AKASHI Takahiro wrote:
>>> In this patch, helper functions for an load option variable (BootXXXX)
>>> are added:
>>> * efi_parse_load_option(): parse a string into load_option data
>>> (renamed from parse_load_option and exported)
>>> * efi_marshel_load_option(): convert load_option data into a string
>>>
>>> Those functions will be used to implement efishell command.
>>>
>>> Signed-off-by: AKASHI Takahiro <takahiro.aka...@linaro.org>
>>> ---
>>> include/efi_loader.h | 25 +++++++++++++
>>> lib/efi_loader/efi_bootmgr.c | 68 ++++++++++++++++++++++++------------
>>> 2 files changed, 70 insertions(+), 23 deletions(-)
>>>
>>> diff --git a/include/efi_loader.h b/include/efi_loader.h
>>> index b73fbb6de23f..1cabb1680d20 100644
>>> --- a/include/efi_loader.h
>>> +++ b/include/efi_loader.h
>>> @@ -502,6 +502,31 @@ efi_status_t EFIAPI efi_set_variable(u16
>>> *variable_name, efi_guid_t *vendor,
>>> u32 attributes, efi_uintn_t data_size,
>>> void *data);
>>>
>>> +/*
>>> + * See section 3.1.3 in the v2.7 UEFI spec for more details on
>>> + * the layout of EFI_LOAD_OPTION. In short it is:
>>> + *
>>> + * typedef struct _EFI_LOAD_OPTION {
>>> + * UINT32 Attributes;
>>> + * UINT16 FilePathListLength;
>>> + * // CHAR16 Description[]; <-- variable length, NULL terminated
>>> + * // EFI_DEVICE_PATH_PROTOCOL FilePathList[];
>>> + * <-- FilePathListLength bytes
>>> + * // UINT8 OptionalData[];
>>> + * } EFI_LOAD_OPTION;
>>> + */
>>> +struct load_option {
>>> + u32 attributes;
>>> + u16 file_path_length;
>>> + u16 *label;
>>> + struct efi_device_path *file_path;
>>> + u8 *optional_data;
>>> +};
>>
>> If this is part of the spec, shouldn't it rather be in efi_api.h?
>
> While uefi spec mentions this structure, I don't have good confidence
> that I can say that it is part of spec or API because there is no interface
> (or function) that handles this structure.
Good point, it's internal only. Then efi_loader.h is probably a good fit.
>
>> It
>> probably also wants an efi_ prefix then :).
>
> OK.
>
>>> +
>>> +void efi_parse_load_option(struct load_option *lo, void *ptr);
>>> +unsigned long efi_marshal_load_option(u32 attr, u16 *label,
>>> + struct efi_device_path *file_path,
>>> + char *option, void **data);
>>> void *efi_bootmgr_load(struct efi_device_path **device_path,
>>> struct efi_device_path **file_path);
>>>
>>> diff --git a/lib/efi_loader/efi_bootmgr.c b/lib/efi_loader/efi_bootmgr.c
>>> index 0c5764db127b..c2d29f956065 100644
>>> --- a/lib/efi_loader/efi_bootmgr.c
>>> +++ b/lib/efi_loader/efi_bootmgr.c
>>> @@ -30,28 +30,8 @@ static const struct efi_runtime_services *rs;
>>> */
>>>
>>>
>>> -/*
>>> - * See section 3.1.3 in the v2.7 UEFI spec for more details on
>>> - * the layout of EFI_LOAD_OPTION. In short it is:
>>> - *
>>> - * typedef struct _EFI_LOAD_OPTION {
>>> - * UINT32 Attributes;
>>> - * UINT16 FilePathListLength;
>>> - * // CHAR16 Description[]; <-- variable length, NULL terminated
>>> - * // EFI_DEVICE_PATH_PROTOCOL FilePathList[]; <--
>>> FilePathListLength bytes
>>> - * // UINT8 OptionalData[];
>>> - * } EFI_LOAD_OPTION;
>>> - */
>>> -struct load_option {
>>> - u32 attributes;
>>> - u16 file_path_length;
>>> - u16 *label;
>>> - struct efi_device_path *file_path;
>>> - u8 *optional_data;
>>> -};
>>> -
>>> /* parse an EFI_LOAD_OPTION, as described above */
>>> -static void parse_load_option(struct load_option *lo, void *ptr)
>>> +void efi_parse_load_option(struct load_option *lo, void *ptr)
>>
>> I think you want to rename this to "deserialize" to better align with ...
>>
>>> {
>>> lo->attributes = *(u32 *)ptr;
>>> ptr += sizeof(u32);
>>> @@ -60,7 +40,7 @@ static void parse_load_option(struct load_option *lo,
>>> void *ptr)
>>> ptr += sizeof(u16);
>>>
>>> lo->label = ptr;
>>> - ptr += (u16_strlen(lo->label) + 1) * 2;
>>> + ptr += (u16_strlen(lo->label) + 1) * sizeof(u16);
>>>
>>> lo->file_path = ptr;
>>> ptr += lo->file_path_length;
>>> @@ -68,6 +48,48 @@ static void parse_load_option(struct load_option *lo,
>>> void *ptr)
>>> lo->optional_data = ptr;
>>> }
>>>
>>> +unsigned long efi_marshal_load_option(u32 attr, u16 *label,
>>
>> ... this function which really should be called "serialize" rather than
>> "marshal". "Marshalling" is what you do to convert from one API to
>> another. Here, we want to write an in-memory object to disk.
>
> Well, as you know, I borrow this word from RPC world.
> While I believe that those two are interchangeable in most contexts,
> I don't care either way if you prefer 'serialize'.
Yeah, I read up on marshalling afterwards too because I got curious and
I guess it really boils down to which world you come from. The Windows
DOM world always calls everything marshalling, while the Java world
calls it serialize usually.
Either way, I personally find serialize the more obvious name :).
>
>> I also think the API would make more sense if you pass in a struct
>> efi_load_option *. It's more symmetric to the deserialize one then.
>
> Absolutely agree.
>
>> You also want to add comments to the functions to say what they do and
>> what their parameters are there for. That will make it easier for people
>> to grasp on how to use this (now public) API.
>
> OK
>
>> And I really dislike void * for anything that isn't "opaque". In this
>> function (and the one above) void * really gets used as byte pointer.
>> Please mark it as such (u8 *).
>
> Right, but there are quite a few of places where "void *" is used
> instead of "u8 *" for a string. For instance, get_var() in bootmgr.c
> or more in other files.
>
> It's quite painful to change all the places for consistency.
Yeah, let's focus on the serialize/deserialize functions for now and
just pass u8* there ;). We can tackle the others when we get to them.
Sorry to make you go through this - I should've caught that back when
Rob submitted the patches.
Alex
>
>> The next problem are the casts. If you cast from u8 * to u32 * and
>> dereference it, the compiler does not know if that pointer is aligned or
>> not. So on platforms that don't enable caches yet in the bootmgr path,
>> we may get unaligned exceptions. So you want to use get_unaligned_le32()
>> and friends or memcpy (as you did in your function) above.
>
> Good point. I will fix them.
> (This reveals other bugs :)
>
> -Takahiro Akashi
>
>>
>> Alex
>>
>>> + struct efi_device_path *file_path,
>>> + char *option, void **data)
>>> +{
>>> + unsigned long size;
>>> + unsigned long label_len, option_len;
>>> + u16 file_path_len;
>>> + void *p;
>>> +
>>> + label_len = (u16_strlen(label) + 1) * sizeof(u16);
>>> + file_path_len = efi_dp_size(file_path)
>>> + + sizeof(struct efi_device_path); /* for END */
>>> + option_len = strlen(option);
>>> +
>>> + /* total size */
>>> + size = sizeof(attr);
>>> + size += file_path_len;
>>> + size += label_len;
>>> + size += option_len + 1;
>>> + p = malloc(size);
>>> + if (!p)
>>> + return 0;
>>> +
>>> + /* copy data */
>>> + *data = p;
>>> + memcpy(p, &attr, sizeof(attr));
>>> + p += sizeof(attr);
>>> + memcpy(p, &file_path_len, sizeof(file_path_len));
>>> + p += sizeof(file_path_len);
>>> + memcpy(p, label, label_len);
>>> + p += label_len;
>>> +
>>> + memcpy(p, file_path, file_path_len);
>>> + p += file_path_len;
>>> +
>>> + memcpy(p, option, option_len);
>>> + p += option_len;
>>> + *(char *)p = '\0';
>>> +
>>> + return size;
>>> +}
>>> +
>>> /* free() the result */
>>> static void *get_var(u16 *name, const efi_guid_t *vendor,
>>> efi_uintn_t *size)
>>> @@ -115,7 +137,7 @@ static void *try_load_entry(uint16_t n, struct
>>> efi_device_path **device_path,
>>> if (!load_option)
>>> return NULL;
>>>
>>> - parse_load_option(&lo, load_option);
>>> + efi_parse_load_option(&lo, load_option);
>>>
>>> if (lo.attributes & LOAD_OPTION_ACTIVE) {
>>> efi_status_t ret;
>>>
_______________________________________________
U-Boot mailing list
U-Boot@lists.denx.de
https://lists.denx.de/listinfo/u-boot
