I was doing some random tests on the b43 hardware RNG.
These are the results.
I patched the firmware to not access the RNG register anymore.
The unpatched firmware does two things. It reads the RNG register to get random values
and it writes 0 to the register every now and then for whatever reason.
Both reads and writes were patched out during my test. So the driver was
the only one accessing the RNG.
This is the result of reading a few bytes from the RNG with the patched fw:
0000000 c4df 4bf4 4d68 d79a 1e5c af3d 7383 5818
0000010 1993 53cc 5ba6 c449 f188 f3c7 e1e6 110b
0000020 8115 f604 4a13 f695 d4db c7a8 0b8e 12e9
0000030 beb8 e4fa 8b37 4217 1c08 8ac6 dc15 148c
0000040 8643 3d91 de0b ea86 53d4 ffa6 2dfb 412b
0000050 f527 6c15 1cd9 d239 d0b4 1a42 9569 622b
0000060 15bf 99d4 5acb 4c4b 9b98 a036 6d41 2490
0000070 393c d7e7 62af a3c5 bdb9 9a0b 8c94 bf30
0000080 eb53 9352 c927 a34c 6847 5cd1 a9b8 5cad
0000090 58f3 53b1 1cb3 f3c7 88e6 4a22 036a bcf9
00000a0 82eb 4efa b8c6 7f8e fa01 47e8 c8e1 fe91
00000b0 9d78 fd74 cc08 d0cc 83a0 f206 23e4 5673
00000c0 d704 9bae ec6f 7726 06ef 04f2 14ee 4329
00000d0 74a0 24e8 19b6 4f67 c6c2 2918 20ad 8483
00000e0 7b00 d4ed 6256 3a77 a28b a945 8953 5827
00000f0 a758 77b0 53dc cd58 006b bafe b474 d6d1
0000100 bdb9 db7a 6f6c f2df 1e6b 6e3d acdc a5a6
0000110 ed60 9349 cf27 c33d 7487 af17 385e 6470
0000120 644b 806f fbfd 1bf6 4b24 d996 2b65 43a9
0000130 2557 48b4 e490 9a93 7634 ead9 e02a 4e04
0000140 fec0 5407 d7a8 9eaf 4a85 a36b 99b9 fbcd
0000150 b38e 2acd cd54 6d65 9191 2cdc 9158 b8bb
0000160 1a74 8fd2 dfe1 7382 46e6 53e7 9ab0 2394
0000170 9981 9232 6c4b d8b1 629c e53a 0369 ef06
0000180 f0c7 fe7d fc03 29f2 4aa7 3429 4697 9a73
0000190 bfd1 cb7f 3f2c 6780 189e 3e30 d583 8b56
00001a0 9579 aeaf 445c 2811 e35d d670 3252 abc8
00001b0 b8be 1e1d 6fc3 18de c69c d172 d8b8 cf9d
00001c0 0596 6df5 804b 5400 8d5c 2c1b e5a7 a26b
00001d0 116f 08bb a123 d0bd 8c43 1519 68a9 332e
00001e0 1ee4 8a79 b3eb ad9b 3d5b 3808 6f8f 6ade
00001f0 8f65 f2e0 5735 75af fad6 210a e542 e769
0000200 af79 2ca1 d959 edc8 ec25 28b1 df50 807c
0000210 83f4 d1f3 585c e19c 3a86 5915 cd4d 929b
0000220 d2de 12b6 8024 7700 0eba 4e1d 0b3a 4d16
0000230 7db4 4604 9b8d 7dc8 2ffb c0bf a403 b56f
0000240 557f f3ab 38e6 7b1e a508 0a94 a914 5d53
0000250 002b 5607 01ac fb06 67f6 fb60 22ee a48b
0000260 6a89 ba57 ed75 3fb4 4680 028d faf7 bff5
0000270 b9f8 b019 799e 791a ebf2 d8a7 16b1 61d2
0000280 fc8b 9bf9 f990 41f2 7182 e9e2 415b c0f8
0000290 1c83 43c7 b386 52cc 92b5 64b4 3f37 da02
00002a0 894f d8b4 8fb0 8ec2 bac4 13eb 7427 5316
00002b0 2703 1562 5ba9 9c49 6f72 4bdf 5f69 4e7d
00002c0 bc7a 4f78 ac60 8b9e fae8 29ea 0f52 55c3
00002d0 3761 65dc 9c96 b639 48b1 a16f 22bd 7289
00002e0 c1cf bb0e ba89 d3eb 0f58 0fc1 923c b1b4
00002f0 6ab0 3dd5 6d16 b7da 52dc faa5 a9eb ea59
0000300 435c 820c e004 8b3f 46a3 e18c d27a 185b
0000310 665f e99a f42c b617 3992 5b18 32b7 a1ca
0000320 1d80 8d76 51e5 b88d 10e2 82de 660a 2033
0000330 3d0a 280b 3d5e 9c08 cbc7 87d1 def1 730f
I'm not sure how random this is, but I expected it to look worse, because I think
the 0 writes in the firmware have an effect on the RNG seeding.
So I expected either all-same values or all-almost-sequencial values.
Didn't happen. That might be a good sign for the quality of the RNG.
So now I did a different test.
Instead of just reading the register I do the following:
- Disable IRQs, which makes the process atomic on this UP machine.
- Write 0 to the RNG
- Read the RNG
- Enable IRQs.
So the write0 -> read sequence should execute in about equal time for each.
This is what happens:
0000000 d547 29b0 ab6f c5bc 252e 9f2e ea07 c454
0000010 b5a2 7443 d358 f437 6fbf 2d0b 9dd8 70ef
0000020 41a5 89ed 766d 3267 f169 908a 4e01 c014
0000030 46c3 9734 7d76 c522 b4a1 b44c 585b 58c7
0000040 2389 6c3d 8636 2765 dac0 0928 4160 fced
0000050 f2ce f3d6 1fc6 5c07 79cc b6c8 089d 4683
0000060 167a ac6f a931 c964 576f 7fba ecf8 b93b
0000070 c157 3810 6fc3 8e0c 44ef 55b2 43ac 92e3
0000080 793e 3a60 7ea7 de18 9013 1a84 e82a 5278
0000090 fb6c 0cbf 0c66 6a3c ca59 dda5 0fdd 967f
00000a0 759e 21ac cf0b 1b7f 2cbd 10c0 d103 7ce9
00000b0 3d80 9411 6bb1 ceb8 fa18 d8a1 51c7 4c10
00000c0 3998 8968 2665 229e 43ef 3fc9 f1fa 1519
00000d0 1501 b452 1fb3 83fd 123f 092c 6bb1 3d48
00000e0 c0eb 7bff 2f4e 0204 c9d6 134d efc7 9ef3
00000f0 98e0 0f84 4100 7fe0 c6fe 5dca 37d1 a27b
0000100 0d75 3e68 b40b 04bc bcb8 7236 182e a92e
0000110 9f09 8a0b 05b1 6043 e2fc 3a2d 2159 7316
0000120 799a c697 ac62 1b3d 3bb9 f126 7f1a 0cfd
0000130 6cb3 9cc1 fb31 b4b3 3e43 130e e263 9e28
0000140 ff20 400b 1303 25c4 6553 2ed2 4e14 c31c
0000150 9857 4171 3bf7 99dd f498 4abd 5652 4e90
0000160 cbc7 d2a2 dfdc cdf1 c2de 7fdb 3903 586d
0000170 e3ba ff3e 3cfa 051d 8d5f b2d9 dadc b1a8
0000180 5f1b 981b 0176 9009 5a7f caa0 0f5e e0f1
0000190 dcfa 3233 dfde 07f3 f689 47b1 c982 4568
00001a0 7582 4e53 7214 e726 5077 7587 82a8 1cd2
00001b0 184b e773 987a a386 071c 3238 6f24 090b
00001c0 f97b 089c e681 0769 283a b577 3f5b b2f6
00001d0 17c9 a341 211b 7213 98dd cd8d c4d0 6122
00001e0 4def 4fd5 3106 eae6 c4ab 93db 4c30 edc0
00001f0 0eb3 e61e ed66 2a94 3fa6 1101 61ee 9018
0000200 2385 9bc9 50bc 9408 4440 5e27 3613 0b6c
0000210 f220 4c25 5d3c d2de 5094 daf2 d4ae 89b4
0000220 48fb de8c 27f0 1f84 eff4 f904 aa9c 6b51
0000230 cd89 aadc a4af 924b b1da d370 3463 c8bf
0000240 a06d 03fe 563f 72b1 8395 09c1 709b 3b09
0000250 151e 0e57 9710 8d85 d3d8 2338 aae7 7aa8
0000260 0a6f 3852 4280 82ea 86d1 b26e 76d9 766e
0000270 90de 7af9 f951 fdcf 3610 8167 41e8 8610
0000280 891e 314f 2d77 7bda 2d4d 0126 2d11 0c91
0000290 5c29 4834 178c e58f 22d0 b112 fc1e c339
00002a0 bd68 72d0 69d6 6e98 db8c 9ebd a3e6 dfc9
00002b0 935d 3963 086f 237f f333 8e3f 6c73 9736
00002c0 2199 2f0c 9cfa 85cd fe5b 64e7 8023 0af9
00002d0 2335 e9e1 0d4a acdb 0933 3f65 b90b fc35
00002e0 2b39 825b 3028 6006 9ff6 3c07 08e7 c6be
00002f0 2144 790f 93cc b0c5 c105 0610 089a 73b8
0000300 762a edb5 7928 ed97 92dc 33dd a364 dc39
0000310 4ff6 6209 a626 84cc f725 4b7f 1b49 7041
0000320 f434 b04f d17d 1be7 3db9 c62e 9898 13c1
0000330 3c32 7dc3 3827 e98d 80b7 6603 6494 1640
I'm not sure about this. There aren't any obvious patterns.
But maybe I'm just blind. Does somebody else see some pattern or
has some RNG test program to recognize such patterns?
So let's do another test. Let's modify the previous test to
write 0xFFFF instead of 0:
0000000 c538 4f9f c9f8 d8b2 b6c7 ff94 78fc ad8e
0000010 b88f 98e4 70c5 620a 012e 45e4 a357 ffff
0000020 e9c6 fc95 60c6 e8f9 1d86 ffff f846 f380
0000030 5bc3 ffff 57ad 5474 f24b 6edb c0ed 7704
0000040 8dce 2393 263d ec67 48c7 55fb 6a5e 1955
0000050 2e98 d4ed 194a d594 76a5 56b0 ee9e cbe3
0000060 8e47 551a afa5 987d d08b c40d 6cb8 5bcd
0000070 76cf 82b2 e8de da72 9a5d 46a8 9330 7e62
0000080 bb33 1543 acaf 0b63 f449 23b9 5e3e 5ce9
0000090 e8ab 2d70 2fdc ffff f4bf f9bd 8392 84c5
00000a0 1335 ffff 1534 95a9 f6af f166 af10 7982
00000b0 6d05 6723 b48a ffff 7a66 2a53 dbaf d2b7
00000c0 8c86 ffff fc5c 3539 8a59 8fa0 08fe ed44
00000d0 61fd c9eb b3b3 3e3e 8d1d a3d2 d136 ffff
00000e0 906b 11fd dbe9 894c bd93 ffff b0eb b601
00000f0 a338 c5c9 902d 0bff 49bd af92 45f3 0fa3
0000100 a794 ffff b31c e332 70c4 a9fa f66d 98b0
0000110 b28a 8769 af7d e7f1 8686 7390 5a61 7f2e
0000120 2ebf 8b19 2741 398a 5b31 15da 3daa 3d24
0000130 b568 b8af 0f76 b3f2 3c31 853e 4ad7 0654
0000140 2ac6 69a3 2a99 94a9 6746 ffff 89e3 b76f
0000150 6ab7 14ad 3440 16bc ffff c4e9 0a42 2d5b
0000160 f466 cfba e507 59ad e09e d902 406b 300d
0000170 6e5c d61c ec60 e298 f9d2 2860 b674 916a
0000180 47f0 7e8b ffff de13 98e8 0f76 2b09 d34d
0000190 afa3 28fe 1c8d 9e18 69e3 ffff 9cf8 213e
00001a0 ffff a624 cf65 9ffb cd02 32d0 7b69 adb9
00001b0 2f10 5906 3296 61dc 93f7 2730 1d7a 092d
00001c0 880b 0d72 e422 01bc ffff 62b3 8edf 3619
00001d0 620f 1428 c95b 1891 7983 4e6b 15ef ffff
00001e0 6ae9 ffff 7c0a d1c8 a8e3 f980 4468 1e26
00001f0 ce75 21e4 f0ec 820d 35d5 9556 c159 ea12
0000200 18df c47d 91b3 a089 630f a633 9560 72a6
0000210 9fcd 4ff6 ba05 16a2 4390 7b19 79b8 e19e
0000220 169a 936d 97c5 b3be f6c3 da96 2e57 c915
0000230 af86 8a7e 1ca6 e73b 0775 1c81 ffff ca58
0000240 5627 ffff 8f98 160d 6f67 87fa 758d 6a50
0000250 eef3 beef 2114 0eea dbe0 8a26 5451 f1bf
0000260 ff62 ffff f709 0388 07cf 1679 2867 ecbd
0000270 fecc c615 cd6b 4bde ffff 5689 1161 1a18
0000280 7cab a136 03e6 7634 3549 3fac 3206 e1d2
0000290 a9d3 be66 aaee 86ab 7064 ffff 3b52 7f48
00002a0 e162 2e19 ee1b ffff 4ac7 d9a9 0061 baf3
00002b0 0657 b19c 87e2 fa75 4bd6 8546 845f b8b7
00002c0 5db5 8cdb 1e3c d717 ffff e63b 9b92 dab9
00002d0 151e c5a4 ffff 6c63 3c39 fd33 a12d 6b16
00002e0 ffff c5c8 4357 663f 5c6f 893f 14b5 30b5
00002f0 9947 2990 ffff b8fa 3280 60d1 64ff 4a47
0000300 ffff 3a96 c552 d0a1 9ef9 911c 8e76 1819
0000310 736c 22c2 2aed 955f cf35 5809 aa7e 1c56
0000320 aa45 ab57 7ab4 b250 ffff 4f13 d3fd abc1
0000330 46eb bc62 2931 616c df0c 2a0d 09a4 d868
Now, that's interesting, isn't it? :)
Similiar stuff happens when writing other magic values.
--
Greetings, Michael.
On Thu, 2009-01-08 at 17:28 -0800, Harvey Harrison wrote:
> On Fri, 2009-01-09 at 02:11 +0100, Michael Buesch wrote:
> > I was doing some random tests on the b43 hardware RNG.
> > These are the results.
> >
> > I patched the firmware to not access the RNG register anymore.
> > The unpatched firmware does two things. It reads the RNG register to get random values
> > and it writes 0 to the register every now and then for whatever reason.
> > Both reads and writes were patched out during my test. So the driver was
> > the only one accessing the RNG.
> >
> > This is the result of reading a few bytes from the RNG with the patched fw:
>
> > I'm not sure about this. There aren't any obvious patterns.
> > But maybe I'm just blind. Does somebody else see some pattern or
> > has some RNG test program to recognize such patterns?
> >
> > So let's do another test. Let's modify the previous test to
> > write 0xFFFF instead of 0:
> >
>
> NIST has a pretty good test suite I think.
Packaged for debian as dieharder if you want the command-line version.
Harvey
On Friday 09 January 2009 02:30:56 Harvey Harrison wrote:
> On Thu, 2009-01-08 at 17:28 -0800, Harvey Harrison wrote:
> > On Fri, 2009-01-09 at 02:11 +0100, Michael Buesch wrote:
> > > I was doing some random tests on the b43 hardware RNG.
> > > These are the results.
> > >
> > > I patched the firmware to not access the RNG register anymore.
> > > The unpatched firmware does two things. It reads the RNG register to get random values
> > > and it writes 0 to the register every now and then for whatever reason.
> > > Both reads and writes were patched out during my test. So the driver was
> > > the only one accessing the RNG.
> > >
> > > This is the result of reading a few bytes from the RNG with the patched fw:
> >
> > > I'm not sure about this. There aren't any obvious patterns.
> > > But maybe I'm just blind. Does somebody else see some pattern or
> > > has some RNG test program to recognize such patterns?
> > >
> > > So let's do another test. Let's modify the previous test to
> > > write 0xFFFF instead of 0:
> > >
> >
> > NIST has a pretty good test suite I think.
>
> Packaged for debian as dieharder if you want the command-line version.
Thanks. I didn't know it was already packaged. So I'll certainly try it.
--
Greetings, Michael.
> I was doing some random tests on the b43 hardware RNG.
> These are the results.
As you're german, you should get a look at the current c't
2/2009. At page 172 it has a in-depth articel about randomness
tests. They also have some tools to download, see
http://von-und-fuer-lau.de/random-tools.html
http://www.heise.de/ct/09/02/links/172.shtml
On Fri, 2009-01-09 at 02:11 +0100, Michael Buesch wrote:
> I was doing some random tests on the b43 hardware RNG.
> These are the results.
>
> I patched the firmware to not access the RNG register anymore.
> The unpatched firmware does two things. It reads the RNG register to get random values
> and it writes 0 to the register every now and then for whatever reason.
> Both reads and writes were patched out during my test. So the driver was
> the only one accessing the RNG.
>
> This is the result of reading a few bytes from the RNG with the patched fw:
> I'm not sure about this. There aren't any obvious patterns.
> But maybe I'm just blind. Does somebody else see some pattern or
> has some RNG test program to recognize such patterns?
>
> So let's do another test. Let's modify the previous test to
> write 0xFFFF instead of 0:
>
NIST has a pretty good test suite I think.
Harvey