[U-Boot] [PATCH v2] common: Fix load and entry addresses in FIT image

[York Sun]

On 09/03/2015 10:55 PM, Simon Glass wrote:
> Hi York,
> 
> On 3 September 2015 at 09:07, York Sun <yorksun at freescale.com> wrote:
>> FIT image supports more than 32 bits in addresses by using #address-cell
>> field. However the address length is not handled when parsing FIT images.
>> Beside, the variable used to host address has "ulong" type. It is OK for
>> the target, but not always enough for host tools such as mkimage. This
>> patch replaces "ulong" with "phys_addr_t" to make sure the address is
>> correct for both the target and the host.
>>
>> Signed-off-by: York Sun <yorksun at freescale.com>
>>
>> ---
>>
>> Changes in v2:
>>   Make a common function for both load and entry addresses.
>>   Simplify calculation of addresses in a similar way as fdtdec_get_number()
>>   fdtdec_get_number() is not used, or too many files need to be included
>>     and/or twisted for host tool
>>   Continue to use %08llx for print format for load and entry addresses
>>     because %pa does not always work for host tool (mkimage)
>>
>>  common/bootm.c     |   13 +++++----
>>  common/image-fit.c |   79 ++++++++++++++++++++++++++++------------------------
>>  include/bootm.h    |    6 ++--
>>  include/image.h    |   12 +++++---
>>  4 files changed, 61 insertions(+), 49 deletions(-)
>>
>> diff --git a/common/bootm.c b/common/bootm.c
>> index 667c934..0672e73 100644
>> --- a/common/bootm.c
>> +++ b/common/bootm.c
>> @@ -325,9 +325,9 @@ static int handle_decomp_error(int comp_type, size_t uncomp_size,
>>         return BOOTM_ERR_RESET;
>>  }
>>
>> -int bootm_decomp_image(int comp, ulong load, ulong image_start, int type,
>> -                      void *load_buf, void *image_buf, ulong image_len,
>> -                      uint unc_len, ulong *load_end)
>> +int bootm_decomp_image(int comp, phys_addr_t load, phys_addr_t image_start,
>> +                      int type, void *load_buf, void *image_buf,
>> +                      ulong image_len, uint unc_len, ulong *load_end)
>>  {
>>         int ret = 0;
>>
>> @@ -883,7 +883,8 @@ void memmove_wd(void *to, void *from, size_t len, ulong chunksz)
>>  static int bootm_host_load_image(const void *fit, int req_image_type)
>>  {
>>         const char *fit_uname_config = NULL;
>> -       ulong data, len;
>> +       phys_addr_t *data = NULL;
>> +       ulong len;
>>         bootm_headers_t images;
>>         int noffset;
>>         ulong load_end;
>> @@ -897,7 +898,7 @@ static int bootm_host_load_image(const void *fit, int req_image_type)
>>         noffset = fit_image_load(&images, (ulong)fit,
>>                 NULL, &fit_uname_config,
>>                 IH_ARCH_DEFAULT, req_image_type, -1,
>> -               FIT_LOAD_IGNORED, &data, &len);
>> +               FIT_LOAD_IGNORED, data, &len);
>>         if (noffset < 0)
>>                 return noffset;
>>         if (fit_image_get_type(fit, noffset, &image_type)) {
>> @@ -912,7 +913,7 @@ static int bootm_host_load_image(const void *fit, int req_image_type)
>>
>>         /* Allow the image to expand by a factor of 4, should be safe */
>>         load_buf = malloc((1 << 20) + len * 4);
>> -       ret = bootm_decomp_image(imape_comp, 0, data, image_type, load_buf,
>> +       ret = bootm_decomp_image(imape_comp, 0, *data, image_type, load_buf,
>>                                  (void *)data, len, CONFIG_SYS_BOOTM_LEN,
>>                                  &load_end);
>>         free(load_buf);
>> diff --git a/common/image-fit.c b/common/image-fit.c
>> index 28f7aa8..7a0c9a9 100644
>> --- a/common/image-fit.c
>> +++ b/common/image-fit.c
>> @@ -358,7 +358,7 @@ void fit_image_print(const void *fit, int image_noffset, const char *p)
>>         char *desc;
>>         uint8_t type, arch, os, comp;
>>         size_t size;
>> -       ulong load, entry;
>> +       phys_addr_t load, entry;
>>         const void *data;
>>         int noffset;
>>         int ndepth;
>> @@ -428,17 +428,17 @@ void fit_image_print(const void *fit, int image_noffset, const char *p)
>>                 if (ret)
>>                         printf("unavailable\n");
>>                 else
>> -                       printf("0x%08lx\n", load);
>> +                       printf("0x%08llx\n", (uint64_t)load);
>>         }
>>
>>         if ((type == IH_TYPE_KERNEL) || (type == IH_TYPE_STANDALONE) ||
>>             (type == IH_TYPE_RAMDISK)) {
>> -               fit_image_get_entry(fit, image_noffset, &entry);
>> +               ret = fit_image_get_entry(fit, image_noffset, &entry);
>>                 printf("%s  Entry Point:  ", p);
>>                 if (ret)
>>                         printf("unavailable\n");
>>                 else
>> -                       printf("0x%08lx\n", entry);
>> +                       printf("0x%08llx\n", (uint64_t)entry);
> 
> Yes of course %pa does not work on the host - I didn't think of that.
> 
> I'm still not thrilled with everything being promoted to 64-bit. Do
> you think using a #define in inttypes.h or similar might work, similar
> to how LBAF works in ide.h?
> 
> #if BITS_PER_LONG == 64
> #define PRIpa "%08l"
> #else
> #define PRIpa "%08l"
> #endif
> 
> The odd thing is that they are both the same for ARM (unsigned long).
> What arch are you using?

This one works for me

#if BITS_PER_LONG == 64
#define PRIpa  "lx"
#else
#define PRIpa  "llx"
#endif

The trick here is host tool. I was testing on armv8.

York