On Tue 2015-09-15 13:02:29 -0400, Kurt Roeckx wrote:
> On Tue, Sep 15, 2015 at 12:53:39PM -0400, Daniel Kahn Gillmor wrote:
>> On Tue 2015-09-15 11:14:17 -0400, Werner Koch wrote:
>> > Good, that works. However, the actual burn_cycles code is more like:
>> >
>> > static int
>> > burn_cycles(unsigned long count)
>> > {
>> > char buffer[16];
>> > return gcry_kdf_derive ("123456789abcdef0", 16,
>> > GCRY_KDF_ITERSALTED_S2K,
>> > GCR_MD_SHA1, "saltsalt", 8,
>> > count, buffer, 16);
>> > }
>> >
>> > and thus does a lot of SHA-1 operations in Libgcrypt.
>>
>> Yep, i wanted the initial test to happen without libgcrypt.
>>
>> Below is a new version that uses gcrypt (fixing a couple typos in
>> Werner's example above). It should be compiled with:
>>
>> gcc -lgcrypt -Wall -Werror --pedantic -o test-csail test-csail.c
>>
>> i'm tracking this test code in git, if anyone is interested:
>> git clone git://lair.fifthhorseman.net/~dkg/789246
>>
>> Kurt, can we try this out on x86-csail-01 ?
>
> S2K calibration: 2 -> 0ms
> S2K calibration: 4 -> 0ms
> S2K calibration: 8 -> 0ms
> S2K calibration: 16 -> 0ms
> S2K calibration: 32 -> 0ms
> S2K calibration: 64 -> 0ms
> S2K calibration: 128 -> 0ms
> S2K calibration: 256 -> 0ms
> S2K calibration: 512 -> 0ms
> S2K calibration: 1024 -> 0ms
> S2K calibration: 2048 -> 0ms
> S2K calibration: 4096 -> 0ms
> S2K calibration: 8192 -> 0ms
> S2K calibration: 16384 -> 0ms
> S2K calibration: 32768 -> 0ms
> S2K calibration: 65536 -> 0ms
> S2K calibration: 131072 -> 0ms
> S2K calibration: 262144 -> 0ms
> S2K calibration: 524288 -> 10ms
> S2K calibration: 1048576 -> 10ms
> S2K calibration: 2097152 -> 20ms
> S2K calibration: 4194304 -> 40ms
> S2K calibration: 8388608 -> 60ms
> S2K calibration: 16777216 -> 120ms
> S2K calibration: 13981013 -> 100ms
> x: 13981013
OK, so that works too. Very weird. Maybe the next thing is to try to
run t-protect.c with:
opt.verbose = 1;
patched into it? I'm a bit at a loss here.
--dkg
