Re: [PATCH v2 08/14] crypto: caam - update rfc4106 sh desc to support zero length input

[Horia Geanta]

On 7/19/2019 2:58 AM, Iuliana Prodan wrote:
> Update share descriptor for rfc4106 to skip instructions in case
> cryptlen is zero. If no instructions are jumped the DECO hangs and a
> timeout error is thrown.
> 
> Signed-off-by: Iuliana Prodan <iuliana.prodan@xxxxxxx>
> ---
>  drivers/crypto/caam/caamalg_desc.c | 46 +++++++++++++++++++++++++-------------
>  drivers/crypto/caam/caamalg_desc.h |  2 +-
>  2 files changed, 31 insertions(+), 17 deletions(-)
> 
> diff --git a/drivers/crypto/caam/caamalg_desc.c b/drivers/crypto/caam/caamalg_desc.c
> index 7253183..99f419a 100644
> --- a/drivers/crypto/caam/caamalg_desc.c
> +++ b/drivers/crypto/caam/caamalg_desc.c
> @@ -843,13 +843,16 @@ EXPORT_SYMBOL(cnstr_shdsc_gcm_decap);
>   * @ivsize: initialization vector size
>   * @icvsize: integrity check value (ICV) size (truncated or full)
>   * @is_qi: true when called from caam/qi
> + *
> + * Input sequence: AAD | PTXT
> + * Output sequence: AAD | CTXT | ICV
> + * AAD length (assoclen), which includes the IV length, is available in Math3.
>   */
>  void cnstr_shdsc_rfc4106_encap(u32 * const desc, struct alginfo *cdata,
>  			       unsigned int ivsize, unsigned int icvsize,
>  			       const bool is_qi)
>  {
> -	u32 *key_jump_cmd;
> -
> +	u32 *key_jump_cmd, *zero_cryptlen_jump_cmd, *skip_instructions;
>  	init_sh_desc(desc, HDR_SHARE_SERIAL);
>  
>  	/* Skip key loading if it is loaded due to sharing */
> @@ -890,26 +893,25 @@ void cnstr_shdsc_rfc4106_encap(u32 * const desc, struct alginfo *cdata,
>  	}
>  
>  	append_math_sub_imm_u32(desc, VARSEQINLEN, REG3, IMM, ivsize);
> -	append_math_add(desc, VARSEQOUTLEN, ZERO, REG3, CAAM_CMD_SZ);
> +	append_math_add(desc, VARSEQOUTLEN, REG0, REG3, CAAM_CMD_SZ);
>  
Why is this needed?
-all math registers are zero when descriptors start execution
-all caam hw revisions have support for math instruction
with ZERO as first operand
-descriptor size stays the same

> -	/* Read assoc data */
> +	/* Skip AAD */
> +	append_seq_fifo_store(desc, 0, FIFOST_TYPE_SKIP | FIFOLDST_VLF);
> +
> +	/* Read cryptlen and set this value into VARSEQOUTLEN */
> +	append_math_sub(desc, VARSEQOUTLEN, SEQINLEN, REG3, CAAM_CMD_SZ);
> +
> +	/* If cryptlen is ZERO jump to AAD command */
> +	zero_cryptlen_jump_cmd = append_jump(desc, JUMP_TEST_ALL |
> +					    JUMP_COND_MATH_Z);
> +
> +	/* Read AAD data */
>  	append_seq_fifo_load(desc, 0, FIFOLD_CLASS_CLASS1 | FIFOLDST_VLF |
>  			     FIFOLD_TYPE_AAD | FIFOLD_TYPE_FLUSH1);
>  
>  	/* Skip IV */
>  	append_seq_fifo_load(desc, ivsize, FIFOLD_CLASS_SKIP);
> -
> -	/* Will read cryptlen bytes */
> -	append_math_sub(desc, VARSEQINLEN, SEQINLEN, REG0, CAAM_CMD_SZ);
> -
> -	/* Workaround for erratum A-005473 (simultaneous SEQ FIFO skips) */
> -	append_seq_fifo_load(desc, 0, FIFOLD_CLASS_CLASS1 | FIFOLD_TYPE_MSG);
> -
Erratum workaround is dropped without any explanation.

The trigger condition (SKIPs both in input and output sequence) still exists.
The descriptor should be validated on a caam with erratum not fixed (era < 6).

Horia