crypto: vmx - Document CTR mode counter width quirks

The CTR code comes from OpenSSL, where it does a 32-bit counter.
The kernel has a 128-bit counter. This difference has lead to
issues.

Document it.

Signed-off-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This commit is contained in:
Daniel Axtens 2019-06-11 11:54:31 +10:00 committed by Herbert Xu
parent 334d37c9e2
commit f651bd96b2

View File

@ -1286,6 +1286,24 @@ ___
#########################################################################
{{{ # CTR procedure[s] #
####################### WARNING: Here be dragons! #######################
#
# This code is written as 'ctr32', based on a 32-bit counter used
# upstream. The kernel does *not* use a 32-bit counter. The kernel uses
# a 128-bit counter.
#
# This leads to subtle changes from the upstream code: the counter
# is incremented with vaddu_q_m rather than vaddu_w_m. This occurs in
# both the bulk (8 blocks at a time) path, and in the individual block
# path. Be aware of this when doing updates.
#
# See:
# 1d4aa0b4c181 ("crypto: vmx - Fixing AES-CTR counter bug")
# 009b30ac7444 ("crypto: vmx - CTR: always increment IV as quadword")
# https://github.com/openssl/openssl/pull/8942
#
#########################################################################
my ($inp,$out,$len,$key,$ivp,$x10,$rounds,$idx)=map("r$_",(3..10));
my ($rndkey0,$rndkey1,$inout,$tmp)= map("v$_",(0..3));
my ($ivec,$inptail,$inpperm,$outhead,$outperm,$outmask,$keyperm,$one)=
@ -1357,7 +1375,7 @@ Loop_ctr32_enc:
addi $idx,$idx,16
bdnz Loop_ctr32_enc
vadduqm $ivec,$ivec,$one
vadduqm $ivec,$ivec,$one # Kernel change for 128-bit
vmr $dat,$inptail
lvx $inptail,0,$inp
addi $inp,$inp,16
@ -1501,7 +1519,7 @@ Load_ctr32_enc_key:
$SHL $len,$len,4
vadduqm $out1,$ivec,$one # counter values ...
vadduqm $out2,$ivec,$two
vadduqm $out2,$ivec,$two # (do all ctr adds as 128-bit)
vxor $out0,$ivec,$rndkey0 # ... xored with rndkey[0]
le?li $idx,8
vadduqm $out3,$out1,$two