ARM v8.3a pauth not working

[Adrien Grassein]

Host: Ubuntu 19.10 - x86_64 machine
QEMU version: 3a63b24a1bbf166e6f455fe43a6bbd8dea413d92 (master)

ARMV8.3 pauth is not working well.

With a test code containing two pauth instructions:
    - paciasp that sign LR with A key and sp as context;
    - autiasp that verify the signature.

Test:
    - Run the program and corrupt LR just before autiasp (ex 0x3e00000400660 instead of 0x3e000000400664)

Expected:
    - autiasp places an invalid pointer in LR

Result:
    - autiasp successfully auth the pointer and places 0x0400660 in LR.

Further explanations:
    Adding traces in qemu code shows that "pauth_computepac" is not robust enough against truncating.
    With 0x31000000400664 as input of pauth_auth, we obtain "0x55b1d65b2c138e14" for PAC, "0x30" for bot_bit and "0x38" for top_bit.
    With 0x310040008743ec as input of pauth (with same key), we obtain "0x55b1d65b2c138ef4" for PAC, "0x30" for bot_bit and "0x38" for top_bit.
    Values of top_bit and bottom_bit are strictly the same and it should not.

[Vincent Dehors (vdehors)]

Hi,

Here is a patch for this bug. The sbox function was using "b+=16" instead of "b+=4".

Also, you check test vector using :

```c
    uint64_t P = 0xfb623599da6e8127ull;
    uint64_t T = 0x477d469dec0b8762ull;
    uint64_t w0 = 0x84be85ce9804e94bull;
    uint64_t k0 = 0xec2802d4e0a488e9ull;
    ARMPACKey key = { .hi = w0, .lo = k0 };
    uint64_t C5 = pauth_computepac(P, T, key);
    /* C5 should be 0xc003b93999b33765 */
```

[Richard Henderson (rth)]

Ooof. Good catch on the sbox error.

That said, how did you test pauth_computepac?
I still do not get the C5 result above, but 0x99d88f4472f3be39.

The following test case sets up the parameters.

[Richard Henderson (rth)]

Oops again. The test case has the parts of the key the wrong way around.
I'll submit the pair of patches to the mailing list.

[Richard Henderson (rth)]

Now upstream as commit de0b1bae6461f67243282555475f88b2384a1eb9.

[Philippe Mathieu-Daudé (philmd)]

Apparently this fixed bug is the official CVE-2020-10702:
https://security-tracker.debian.org/tracker/CVE-2020-10702