Re: [PATCH 1/1] x86: Change mov $0, %reg with xor %reg, %reg

[Kanna Scarlet <knscarlet@gnuweeb.org>]

On 8/4/22 10:53 PM, Borislav Petkov wrote:
> Bonus points if you find out what other advantage
>
> XOR reg,reg
>
> has when it comes to clearing integer registers.

Hello sir Borislav,

Thank you for your response. I tried to find out other advantages of
xor reg,reg on Google and found this:
https://stackoverflow.com/a/33668295/7275114

  "xor (being a recognized zeroing idiom, unlike mov reg, 0) has some
  obvious and some subtle advantages:

  1. smaller code-size than mov reg,0. (All CPUs)
  2. avoids partial-register penalties for later code.
     (Intel P6-family and SnB-family).
  3. doesn't use an execution unit, saving power and freeing up
     execution resources. (Intel SnB-family)
  4. smaller uop (no immediate data) leaves room in the uop cache-line
     for nearby instructions to borrow if needed. (Intel SnB-family).
  5. doesn't use up entries in the physical register file. (Intel
     SnB-family (and P4) at least, possibly AMD as well since they use
     a similar PRF design instead of keeping register state in the ROB
     like Intel P6-family microarchitectures.)"

Should I add all in the explanation sir? I will send v2 revision
tomorrow.

We also find more files to patch with this command:

   grep -rE "mov.?\s+\\$\\0\s*," arch/x86

it shows many immediate zero moves to 64-bit register in file
arch/x86/crypto/curve25519-x86_64.c, but the next instruction may depend
on the previous %rflags value, we are afraid to change this because
xor touches %rflags. We will try to change it to movl $0, %r32 to
reduce the code size.

Example cmovc needs %rflags

    "  adcx %1, %%r11;"
    "  movq %%r11, 24(%2);"

    /* Step 3: Fold the carry bit back in; guaranteed not to carry at this point */
    "  mov $0, %%rax;"
    "  cmovc %%rdx, %%rax;"

Thanks.

Regards,
-- 
Kanna Scarlet