Re: [PATCH] hw/loongarch/boot: Support Linux raw boot image

[Jiaxun Yang]

在2024年12月24日十二月 上午4:03，bibo mao写道：
> On 2024/12/23 上午8:30, Jiaxun Yang wrote:
>> Many distros are shipping raw kernel images (i.e. vmlinux.efi).
>> 
>> Support booting such image by parsing header as per Linux's
>> specification [1].
>> 
>> [1]: https://docs.kernel.org/arch/loongarch/booting.html
>> 
>> Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
>> ---
>> It is based on my previous booting protocol patch
>> ---
>>   hw/loongarch/boot.c         | 45 
>> +++++++++++++++++++++++++++++++++++++++++++++
>>   include/hw/loongarch/boot.h | 17 +++++++++++++++++
>>   2 files changed, 62 insertions(+)
>> 
>> diff --git a/hw/loongarch/boot.c b/hw/loongarch/boot.c
>> index 
>> 93847b0eaf8e50ce1a990b91267780e6785e1c2f..5bc889c51fafa9c6d37426b9bee9709c12183927
>>  100644
>> --- a/hw/loongarch/boot.c
>> +++ b/hw/loongarch/boot.c
>> @@ -260,6 +260,43 @@ static uint64_t cpu_loongarch_virt_to_phys(void 
>> *opaque, uint64_t addr)
>>       return addr & MAKE_64BIT_MASK(0, TARGET_PHYS_ADDR_SPACE_BITS);
>>   }
>>   
>> +static int64_t get_linux_image_info(struct loongarch_boot_info *info,
>> +                                    uint64_t *kernel_entry,
>> +                                    uint64_t *kernel_low,
>> +                                    uint64_t *kernel_high)
>> +{
>> +    int fd;
>> +    struct loongarch_linux_hdr hdr;
>> +    int64_t kernel_size = -1;
>> +
>> +    fd = open(info->kernel_filename, O_RDONLY | O_BINARY);
>> +    if (fd < 0) {
>> +        return -1;
>> +    }
>> +
>> +    if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr)) {
>> +        close(fd);
>> +        return -1;
>> +    }
>> +
>> +    if ((le32_to_cpu(hdr.mz_magic) & 0xffff) != MZ_MAGIC ||
>> +        le32_to_cpu(hdr.linux_pe_magic) != LINUX_PE_MAGIC) {
>> +        close(fd);
>> +        return -1;
>> +    }
>> +
>> +    *kernel_entry = le64_to_cpu(hdr.kernel_entry);
>> +    /* Early kernel versions may have those fields in virtual address */
>> +    *kernel_entry &= MAKE_64BIT_MASK(0, TARGET_PHYS_ADDR_SPACE_BITS);
>> +    *kernel_low = le64_to_cpu(hdr.load_offset);
>> +    *kernel_low &= MAKE_64BIT_MASK(0, TARGET_PHYS_ADDR_SPACE_BITS);
>> +    kernel_size = le64_to_cpu(hdr.kernel_size);
>> +    *kernel_high = *kernel_low + kernel_size;
>> +
>> +    close(fd);
>> +    return kernel_size;
>> +}
>> +
>>   static int64_t load_kernel_info(struct loongarch_boot_info *info)
>>   {
>>       uint64_t kernel_entry, kernel_low, kernel_high;
>> @@ -270,6 +307,14 @@ static int64_t load_kernel_info(struct 
>> loongarch_boot_info *info)
>>                              &kernel_entry, &kernel_low,
>>                              &kernel_high, NULL, 0,
>>                              EM_LOONGARCH, 1, 0);
>> +    if (kernel_size < 0) {
>> +        kernel_size = get_linux_image_info(info, &kernel_entry,
>> +                                           &kernel_low, &kernel_high);
>> +        if (kernel_size >= 0) {
>> +            kernel_size = load_image_targphys(info->kernel_filename,
>> +                                              kernel_low, kernel_size);
>> +        }
> Can we read file header file firstly?
> If it is ELFMAG, load_elf is called.
>
> If it is EFI_PE_MSDOS_MAGIC, check zimg magic in linux_efi_zboot_header, 
> else loongarch_linux_hdr for plain image if zimg does not match.

I'd say as checking will be done by load_elf, the fallback path makes more
sense here.

See similar design on Arm's arm_setup_direct_kernel_boot.

But you are right, I should handle vmlinuz.efi with linux_efi_zboot_header
as well.

>
> Regards
> Bibo Mao
>
>> +    }
[...]

-- 
- Jiaxun