Hardware AES / security system fails on H3

zipob · Published in 2015-10-20 06:28:27

The driver supplied with kernel 3.4.39 for hardware crypto security system (ss) fails with large AES requests, so it's unusable. What does "Too many scatter" even mean? The new sunxi-ss driver in mainline kernel works fine with A20, but there is no support for H3 in it nor is H3 usable with mainline kernel yet for that matter.

Can anything be done or do we have to wait for mainline kernel and sunxi-ss to get support for H3?

cryptsetup benchmark fails with ss module loaded:

root@OrangePI:~# cryptsetup benchmark
# Tests are approximate using memory only (no storage IO).
PBKDF2-sha1 137970 iterations per second
PBKDF2-sha256 93622 iterations per second
PBKDF2-sha512 42555 iterations per second
PBKDF2-ripemd160 126030 iterations per second
PBKDF2-whirlpool 10336 iterations per second
# Algorithm | Key | Encryption | Decryption
aes-cbc 128b N/A N/A
serpent-cbc 128b N/A N/A
twofish-cbc 128b N/A N/A
aes-cbc 256b N/A N/A
serpent-cbc 256b N/A N/A
twofish-cbc 256b N/A N/A
aes-xts 256b N/A N/A
serpent-xts 256b N/A N/A
twofish-xts 256b N/A N/A
aes-xts 512b N/A N/A
serpent-xts 512b N/A N/A
twofish-xts 512b N/A N/A

Copy code

logged output from af_alg_test shortened with uniq -c

root@OrangePI:~# uniq -c log
1 root@OrangePI:~# ./af_alg_test aes
1 INFO: ./af_alg_test will do 1000.000000 request
1 aes 1000.000000 requests of 16 in 126534.000000us (0.126534s) 0.007903r/us 7.903014r/ms 7903.014160r/s
1 aes 1000.000000 requests of 32 in 123696.000000us (0.123696s) 0.008084r/us 8.084335r/ms 8084.335938r/s
1 aes 1000.000000 requests of 64 in 133176.000000us (0.133176s) 0.007509r/us 7.508861r/ms 7508.860352r/s
1 aes 1000.000000 requests of 128 in 132345.000000us (0.132345s) 0.007556r/us 7.556009r/ms 7556.008789r/s
1 aes 1000.000000 requests of 256 in 133477.000000us (0.133477s) 0.007492r/us 7.491928r/ms 7491.927246r/s
1 aes 1000.000000 requests of 512 in 146131.000000us (0.146131s) 0.006843r/us 6.843175r/ms 6843.174805r/s
1 aes 1000.000000 requests of 1024 in 150066.000000us (0.150066s) 0.006664r/us 6.663734r/ms 6663.734863r/s
1 aes 1000.000000 requests of 2048 in 164501.000000us (0.164501s) 0.006079r/us 6.078990r/ms 6078.990234r/s
1 aes 1000.000000 requests of 4096 in 214635.000000us (0.214635s) 0.004659r/us 4.659072r/ms 4659.072266r/s
1 aes 1000.000000 requests of 8192 in 370977.000000us (0.370977s) 0.002696r/us 2.695585r/ms 2695.584961r/s
1 aes 1000.000000 requests of 16384 in 576595.000000us (0.576595s) 0.001734r/us 1.734320r/ms 1734.319580r/s
1000 Did not receive the right amount of data
1 aes 1000.000000 requests of 32768 in 974193.000000us (0.974193s) 0.001026r/us 1.026491r/ms 1026.490601r/s
1000 Did not receive the right amount of data
1 aes 1000.000000 requests of 65536 in 1093468.000000us (1.093468s) 0.000915r/us 0.914522r/ms 914.521484r/s
100 Did not receive the right amount of data
1 aes 100.000000 requests of 131072 in 96782.000000us (0.096782s) 0.001033r/us 1.033250r/ms 1033.250000r/s
100 Did not receive the right amount of data
1 aes 100.000000 requests of 262144 in 104389.000000us (0.104389s) 0.000958r/us 0.957955r/ms 957.955322r/s
100 Did not receive the right amount of data
1 aes 100.000000 requests of 524288 in 94312.000000us (0.094312s) 0.001060r/us 1.060310r/ms 1060.310425r/s
100 Did not receive the right amount of data
1 aes 100.000000 requests of 1048576 in 107692.000000us (0.107692s) 0.000929r/us 0.928574r/ms 928.574097r/s
100 Did not receive the right amount of data
1 aes 100.000000 requests of 2097152 in 96809.000000us (0.096809s) 0.001033r/us 1.032962r/ms 1032.961792r/s

Copy code

fault logged to dmesg

[1311385.175080] ------------[ cut here ]------------
[1311385.175096] WARNING: at drivers/crypto/sunxi/v3/sunxi_ss_proc.c:101 ss_sg_config+0x5c/0x1d8 [ss]()
[1311385.175105] Too many scatter: 8
[1311385.175110]
Modules linked in: ss crypto_user algif_hash algif_skcipher af_alg
snd_usb_audio snd_usbmidi_lib snd_hwdep uvcvideo [last unloaded: ss]
[1311385.175152] [<c0016a6c>] (unwind_backtrace+0x0/0xec) from [<c06a4bd8>] (dump_stack+0x20/0x24)
[1311385.175167] [<c06a4bd8>] (dump_stack+0x20/0x24) from [<c0027a18>] (warn_slowpath_common+0x5c/0x74)
[1311385.175181] [<c0027a18>] (warn_slowpath_common+0x5c/0x74) from [<c0027aec>] (warn_slowpath_fmt+0x40/0x48)
[1311385.175202] [<c0027aec>] (warn_slowpath_fmt+0x40/0x48)rom [<bf08611c>] (ss_sg_config+0x5c/0x1d8 [ss])
[1311385.175229] [<bf08611c>] (ss_sg_config+0x5c/0x1d8 [ss]) from [<bf086b8c>] (ss_aes_start+0x4f0/0x934 [ss])
[1311385.175258] [<bf086b8c>] (ss_aes_start+0x4f0/0x934 [ss]) from [<bf087b84>] (ss_aes_one_req+0x13c/0x1a8 [ss])
[1311385.175285] [<bf087b84>] (ss_aes_one_req+0x13c/0x1a8 [ss]) from [<bf085888>] (sunxi_ss_work+0x94/0x170 [ss])
[1311385.175307] [<bf085888>] (sunxi_ss_work+0x94/0x170 [ss]) from [<c0043108>] (process_one_work+0x294/0x48c)
[1311385.175322] [<c0043108>] (process_one_work+0x294/0x48c) from [<c0043674>] (worker_thread+0x1a8/0x2a4)
[1311385.175337] [<c0043674>] (worker_thread+0x1a8/0x2a4) from [<c004837c>] (kthread+0x9c/0xa8)
[1311385.175352] [<c004837c>] (kthread+0x9c/0xa8) from [<c000f360>] (kernel_thread_exit+0x0/0x8)
[1311385.175361] ---[ end trace 25082aad9cb13992 ]---
[1311385.175396] ss_aes_start()372 - CE return error: 2
[1311385.175403] ss_aes_one_req()685 - ss_aes_start fail(-22)

Copy code

igorpec · Published in 2015-10-22 00:55:01

Edited by igorpec at 2015-10-21 17:57

Can anything be done or do we have to wait for mainline kernel and sunxi-ss to get support for H3?

http://sunxi.montjoie.ovh/
sunxi-ss also doesn't work properly on A20 with kernel 3.4 and I think author is not dealing with kernel 3.4 any more ... forget about it until H3 reaches mainline.

zipob · Published in 2015-12-28 04:06:58

Github user omonar seems to have patched the driver. https://github.com/omonar/OrangePI-Kernel/commit/26bb10ea978ba61e582f69f91dfef628fafd8c80

There is noticeable increase in aes-cbc performance.

cryptsetup benchmark without ss module loaded

# Algorithm | Key | Encryption | Decryption
aes-cbc 128b 19.1 MiB/s 19.8 MiB/s
aes-cbc 256b 14.7 MiB/s 15.0 MiB/s

Copy code

with ss loaded

# Algorithm | Key | Encryption | Decryption
aes-cbc 128b 37.7 MiB/s 37.8 MiB/s
aes-cbc 256b 33.0 MiB/s 32.6 MiB/s

Copy code