Fix some potential unaligned accesses in dtc

Because of the convention of packed representations in property layouts,
it's not uncommon to have integer values in properties which aren't
naturally aligned.  Thus, there are several places in the dtc code where we
cast a potentially unaligned byte pointer into an integer pointer and load
it directly.  On a number of architectures (including sparc64 and arm) this
won't work and will cause a fault.  In some cases it may be trapped and
emulated by the kernel, but not always.

Therefore, replace such direct unaligned reads with a helper which will
handle unaligned data reads (a variant on the fdtXX_ld() functions already
used in libfdt).

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This commit is contained in:
David Gibson 2020-04-14 15:02:51 +10:00
parent 87a656ae5f
commit b28464a550
4 changed files with 38 additions and 7 deletions

View file

@ -110,13 +110,13 @@ static void write_propval_int(FILE *f, const char *p, size_t len, size_t width)
fprintf(f, "%02"PRIx8, *(const uint8_t*)p);
break;
case 2:
fprintf(f, "0x%02"PRIx16, fdt16_to_cpu(*(const fdt16_t*)p));
fprintf(f, "0x%02"PRIx16, dtb_ld16(p));
break;
case 4:
fprintf(f, "0x%02"PRIx32, fdt32_to_cpu(*(const fdt32_t*)p));
fprintf(f, "0x%02"PRIx32, dtb_ld32(p));
break;
case 8:
fprintf(f, "0x%02"PRIx64, fdt64_to_cpu(*(const fdt64_t*)p));
fprintf(f, "0x%02"PRIx64, dtb_ld64(p));
break;
}
if (p + width < end)