Re: HRNG in CAAM isn't working properly on IMX6 SoloX

[Fabio Estevam]

Hi Fredrik,

On Mon, Aug 30, 2021 at 8:22 AM Fredrik Yhlen <fredrik.yhlen@xxxxxxxxx> wrote:
>
> Hi,
>
> We're having problems with hwrng on a board with imx6sx (soloX) running Linux
> 5.10.x. mainline, and I have tracked it down to this commit
> '358ba762d9f1d4ba99ab31ef12bc28014b22f4c9' as being the culprit.
>
> The caam_jr driver spits out lots of messages when attempting to read from /dev/hwrng:
> ```
> [29717.629041] hwrng: no data available
> [29727.859008] caam_jr 2101000.jr: 20003c5b: CCB: desc idx 60: RNG: Hardware error
> ```
>
> ```
> caam_jr 2101000.jr0: 2000025b: CCB: desc idx 2: RNG: Hardware error.
> caam_jr 2101000.jr0: 20003c5b: CCB: desc idx 60: RNG: Hardware error.
> caam_jr 2101000.jr0: 20003c5b: CCB: desc idx 60: RNG: Hardware error.
> caam_jr 2101000.jr0: 20003c5b: CCB: desc idx 60: RNG: Hardware error.
> caam_jr 2101000.jr0: 20003c5b: CCB: desc idx 60: RNG: Hardware error.
> caam_jr 2101000.jr0: 20003c5b: CCB: desc idx 60: RNG: Hardware error.
> ```
>
> This also happens on Boundary's Nitrogen6_soloX board when running the same
> kernel, and likewise with their latest Yocto release that uses 5.4.100 linux-imx kernel.
>
> ```
> root@nitrogen6sx:~# dd if=/dev/hwrng of=/tmp/random bs=256 count=10
> [  113.940735] caam_jr 2101000.jr0: 20003c5b: CCB: desc idx 60: RNG: Hardware error
> dd: /dev/hwrng: Invalid argument
> root@nitrogen6sx:~# rm /tmp/random
> root@nitrogen6sx:~# dd if=/dev/hwrng of=/tmp/random bs=256 count=10
> [  125.300823] caam_jr 2101000.jr0: 20003c5b: CCB: desc idx 60: RNG: Hardware error
> dd: /dev/hwrng: Invalid argument
> root@nitrogen6sx:~# du -hs /tmp/random
> 0       /tmp/random
> root@nitrogen6sx:~# ls -l /tmp/random
> -rw-r--r--    1 root     root             0 Dec 16 17:27 /tmp/random
> root@nitrogen6sx:~#
> ```
>
> And then no data is available from /dev/hwrng.
>
> The problem occurs when adding OP_ALG_PR_ON(prediction resistance) when setting up
> job descriptor for reading new random data in caamrng.c. There are also
> some confusing parts about this commit that I'm not too sure about.
>
> 1. It's adding a conditional variable named 'pr_support', but I guess this only
> indicates if the MC(Management Complex) supports prediction resistance,
> since the following check can be bypassed when 'pr_support' is false.
>
>     /*
>      * If SEC has RNG version >= 4 and RNG state handle has not been
>      * already instantiated, do RNG instantiation
>      * In case of SoCs with Management Complex, RNG is managed by MC f/w.
>      */
>     if (!(ctrlpriv->mc_en && pr_support) && rng_vid >= 4) {
>
>
> This will eventually lead to the following chain call: caam_probe() -> instantiate_rng() ->
> build_instantiation_desc(), where OP_ALG_PR_ON will be used through DECO.
>
> static void build_instantiation_desc(u32 *desc, int handle, int do_sk)
> {
>     u32 *jump_cmd, op_flags;
>
>     init_job_desc(desc, 0);
>
>     op_flags = OP_TYPE_CLASS1_ALG | OP_ALG_ALGSEL_RNG |
>             (handle << OP_ALG_AAI_SHIFT) | OP_ALG_AS_INIT |
>             OP_ALG_PR_ON;
>     ...
>     ...
>     ...
>
> Shouldn't it be named 'mc_pr_support' instead, or something similar?
>
> 2. PR is unconditionally used in caamrng.c(caam_jr module) when
> reading new RNG data. Should this be the case?
>
> Removing OP_ALG_PR_ON in caam_init_desc() from drivers/crypto/caam/caamrng.c
> seems to fix the problem we're experiencing, here's an example:
> ```
> diff --git a/drivers/crypto/caam/caamrng.c b/drivers/crypto/caam/caamrng.c
> index 77d048dfe5d0..f085a80b1b3c 100644
> --- a/drivers/crypto/caam/caamrng.c
> +++ b/drivers/crypto/caam/caamrng.c
> @@ -67,8 +67,7 @@ static u32 *caam_init_desc(u32 *desc, dma_addr_t dst_dma)
>  {
>         init_job_desc(desc, 0); /* + 1 cmd_sz */
>         /* Generate random bytes: + 1 cmd_sz */
> -       append_operation(desc, OP_ALG_ALGSEL_RNG | OP_TYPE_CLASS1_ALG |
> -                        OP_ALG_PR_ON);
> +       append_operation(desc, OP_ALG_ALGSEL_RNG | OP_TYPE_CLASS1_ALG);
>         /* Store bytes: + 1 cmd_sz + caam_ptr_sz  */
>         append_fifo_store(desc, dst_dma,
>                           CAAM_RNG_MAX_FIFO_STORE_SIZE, FIFOST_TYPE_RNGSTORE);
> ```

We also observe this issue.

Heiko on Cc tried increasing the RTSDCTL_ENT_DLY_MIN value
and this seems to help:

diff --git a/drivers/crypto/caam/regs.h b/drivers/crypto/caam/regs.h
index af61f3a2c0d4..53c9fa04a24c 100644
--- a/drivers/crypto/caam/regs.h
+++ b/drivers/crypto/caam/regs.h
@@ -513,7 +513,7 @@ struct rng4tst {
  };
 #define RTSDCTL_ENT_DLY_SHIFT 16
 #define RTSDCTL_ENT_DLY_MASK (0xffff << RTSDCTL_ENT_DLY_SHIFT)
-#define RTSDCTL_ENT_DLY_MIN 3200
+#define RTSDCTL_ENT_DLY_MIN 4800
 #define RTSDCTL_ENT_DLY_MAX 12800
  u32 rtsdctl; /* seed control register */
  union {

Does this help in your case?

Thanks,

Fabio Estevam