Hi Stephen,
On 08/24/2012 02:13 AM, Lukasz Majewski wrote:
Custom definitions of le_XX_to_int functions have been replaced with standard ones, defined at <compiler.h>
diff --git a/disk/part_efi.c b/disk/part_efi.c
-/* Convert char[8] in little endian format to the host format integer
- */
-static inline unsigned long long le64_to_int(unsigned char *le64) -{
So this original function takes a pointer to the value ...
/* Check the first_usable_lba and last_usable_lba are within the disk. */ lastlba = (unsigned long long)dev_desc->lba;
- if (le64_to_int(pgpt_head->first_usable_lba) > lastlba) {
- if (le64_to_cpu(pgpt_head->first_usable_lba) > lastlba) {
At this point in the series, first_usable_lba is a char[8], so this is passing the address of the first byte to both the original function le64_to_int(), and the replacement function le64_to_cpu(). However, le64_to_cpu() expects to receive the 64-bit value to swap, not a pointer to it - from compiler.h:
#define _uswap_64(x, sfx) \ ((((x) & 0xff00000000000000##sfx) >> 56) | \ ... # define uswap_64(x) _uswap_64(x, ull)
le: # define le64_to_cpu(x) (x) be: # define le64_to_cpu(x) uswap_64(x)
So I think this patch breaks the code, and then the next patch fixes it, since it changes the type of first_usable_lba from char[8] to __le64?
First of all, thanks for very thorough review.
You are right, my fault. Patch 2/6 and 3/6 shall be squashed together. When squashed, no errors and warnings appear.
I've unnecessary separated them to show this two step change. For the sake of bisecting - those shall be squashed.