Re: XTS Crypto Not Found In /proc/crypto Even After Compiled for 4.10.1.

[Herbert Xu]

On Thu, Mar 02, 2017 at 05:35:30PM -0600, Nathan Royce wrote:
> ARM ODroid XU4
> 
> $ cat /proc/config.gz | gunzip | grep XTS
> CONFIG_CRYPTO_XTS=y
> 
> $ grep xts /proc/crypto
> //4.9.13
> name         : xts(aes)
> driver       : xts(aes-generic)
> //4.10.1
> <blank>
> //cbc can be found though
> 
> CRYPTTAB:
> cryptswap1 UUID=<sanitized> /dev/urandom
> swap,offset=2048,cipher=aes-xts-plain64:sha512,size=512,nofail
> 
> FSTAB:
> /dev/mapper/cryptswap1 none swap sw 0 0
> 
> Boot Log:
> [   10.535985] ------------[ cut here ]------------
> [   10.539252] WARNING: CPU: 0 PID: 0 at crypto/skcipher.c:430
> skcipher_walk_first+0x13c/0x14c
> [   10.547542] Modules linked in: xor xor_neon aes_arm zlib_deflate
> dm_crypt raid6_pq nfsd auth_rpcgss oid_registry nfs_acl lockd grace sunrpc
> ip_tables x_tables
> [   10.561716] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.10.1-dirty #1
> [   10.568049] Hardware name: SAMSUNG EXYNOS (Flattened Device Tree)
> [   10.574171] [<c010eb54>] (unwind_backtrace) from [<c010b408>]
> (show_stack+0x10/0x14)
> [   10.581893] [<c010b408>] (show_stack) from [<c036772c>]
> (dump_stack+0x84/0x98)
> [   10.589073] [<c036772c>] (dump_stack) from [<c011bb20>]
> (__warn+0xe8/0x100)
> [   10.595975] [<c011bb20>] (__warn) from [<c011bc30>]
> (warn_slowpath_null+0x20/0x28)
> [   10.603546] [<c011bc30>] (warn_slowpath_null) from [<c0329a48>]
> (skcipher_walk_first+0x13c/0x14c)
> [   10.612390] [<c0329a48>] (skcipher_walk_first) from [<c0329b30>]
> (skcipher_walk_virt+0x1c/0x38)
> [   10.621056] [<c0329b30>] (skcipher_walk_virt) from [<c0330ed0>]
> (post_crypt+0x38/0x1c4)
> [   10.629022] [<c0330ed0>] (post_crypt) from [<c0331470>]
> (decrypt_done+0x4c/0x54)
> [   10.636389] [<c0331470>] (decrypt_done) from [<c05a03f0>]
> (s5p_aes_complete+0x70/0xfc)
> [   10.644274] [<c05a03f0>] (s5p_aes_complete) from [<c05a05b0>]
> (s5p_aes_interrupt+0x134/0x1a0)
> [   10.652771] [<c05a05b0>] (s5p_aes_interrupt) from [<c016dc3c>]
> (__handle_irq_event_percpu+0x9c/0x124)

This looks like a bug in the s5p driver.  It's calling the completion
function straight from the IRQ handler, which is triggering the
sanity check in skcipher_walk_first.

The s5p driver needs to schedule a tasklet to call the completion
function.

Do you have crypto self-test enabled? If so it should've caught
this at run-time.  Otherwise you can disable the s5p driver until
it's fixed.

Cheers,
-- 
Email: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt