RE: [PATCH] crypto: caam - optimize RNG sample size

[Gaurav Jain]

Reviewed-by: Gaurav Jain <gaurav.jain@xxxxxxx>

> -----Original Message-----
> From: Meenakshi Aggarwal <meenakshi.aggarwal@xxxxxxx>
> Sent: Monday, June 12, 2023 2:01 PM
> To: Horia Geanta <horia.geanta@xxxxxxx>; Varun Sethi <V.Sethi@xxxxxxx>;
> Pankaj Gupta <pankaj.gupta@xxxxxxx>; Gaurav Jain <gaurav.jain@xxxxxxx>;
> herbert@xxxxxxxxxxxxxxxxxxx; davem@xxxxxxxxxxxxx; linux-
> crypto@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx
> Cc: Meenakshi Aggarwal <meenakshi.aggarwal@xxxxxxx>
> Subject: [PATCH] crypto: caam - optimize RNG sample size
> 
> From: Meenakshi Aggarwal <meenakshi.aggarwal@xxxxxxx>
> 
> TRNG "sample size" (the total number of entropy samples that will be taken
> during entropy generation) default / POR value is very conservatively set to 2500.
> 
> Let's set it to 512, the same as the caam driver in U-boot
> (drivers/crypto/fsl_caam.c) does.
> 
> This solves the issue of RNG performance dropping after a suspend/resume
> cycle on parts where caam loses power, since the initial U-boot setttings are lost
> and kernel does not restore them when resuming.
> 
> Note: when changing the sample size, the self-test parameters need to be
> updated accordingly.
> 
> Signed-off-by: Horia Geantă <horia.geanta@xxxxxxx>
> Signed-off-by: Meenakshi Aggarwal <meenakshi.aggarwal@xxxxxxx>
> ---
>  drivers/crypto/caam/ctrl.c | 52 +++++++++++++++++++++++---------------
>  drivers/crypto/caam/regs.h | 14 ++++++++--
>  2 files changed, 44 insertions(+), 22 deletions(-)
> 
> diff --git a/drivers/crypto/caam/ctrl.c b/drivers/crypto/caam/ctrl.c index
> 62dd069942e4..b06bb64c6c23 100644
> --- a/drivers/crypto/caam/ctrl.c
> +++ b/drivers/crypto/caam/ctrl.c
> @@ -352,7 +352,7 @@ static void kick_trng(struct device *dev, int ent_delay)
>  	struct caam_drv_private *ctrlpriv = dev_get_drvdata(dev);
>  	struct caam_ctrl __iomem *ctrl;
>  	struct rng4tst __iomem *r4tst;
> -	u32 val;
> +	u32 val, rtsdctl;
> 
>  	ctrl = (struct caam_ctrl __iomem *)ctrlpriv->ctrl;
>  	r4tst = &ctrl->r4tst[0];
> @@ -368,26 +368,38 @@ static void kick_trng(struct device *dev, int ent_delay)
>  	 * Performance-wise, it does not make sense to
>  	 * set the delay to a value that is lower
>  	 * than the last one that worked (i.e. the state handles
> -	 * were instantiated properly. Thus, instead of wasting
> -	 * time trying to set the values controlling the sample
> -	 * frequency, the function simply returns.
> +	 * were instantiated properly).
>  	 */
> -	val = (rd_reg32(&r4tst->rtsdctl) & RTSDCTL_ENT_DLY_MASK)
> -	      >> RTSDCTL_ENT_DLY_SHIFT;
> -	if (ent_delay <= val)
> -		goto start_rng;
> -
> -	val = rd_reg32(&r4tst->rtsdctl);
> -	val = (val & ~RTSDCTL_ENT_DLY_MASK) |
> -	      (ent_delay << RTSDCTL_ENT_DLY_SHIFT);
> -	wr_reg32(&r4tst->rtsdctl, val);
> -	/* min. freq. count, equal to 1/4 of the entropy sample length */
> -	wr_reg32(&r4tst->rtfrqmin, ent_delay >> 2);
> -	/* max. freq. count, equal to 16 times the entropy sample length */
> -	wr_reg32(&r4tst->rtfrqmax, ent_delay << 4);
> -	/* read the control register */
> -	val = rd_reg32(&r4tst->rtmctl);
> -start_rng:
> +	rtsdctl = rd_reg32(&r4tst->rtsdctl);
> +	val = (rtsdctl & RTSDCTL_ENT_DLY_MASK) >> RTSDCTL_ENT_DLY_SHIFT;
> +	if (ent_delay > val) {
> +		val = ent_delay;
> +		/* min. freq. count, equal to 1/4 of the entropy sample length
> */
> +		wr_reg32(&r4tst->rtfrqmin, val >> 2);
> +		/* max. freq. count, equal to 16 times the entropy sample
> length */
> +		wr_reg32(&r4tst->rtfrqmax, val << 4);
> +	}
> +
> +	wr_reg32(&r4tst->rtsdctl, (val << RTSDCTL_ENT_DLY_SHIFT) |
> +		 RTSDCTL_SAMP_SIZE_VAL);
> +
> +	/*
> +	 * To avoid reprogramming the self-test parameters over and over again,
> +	 * use RTSDCTL[SAMP_SIZE] as an indicator.
> +	 */
> +	if ((rtsdctl & RTSDCTL_SAMP_SIZE_MASK) != RTSDCTL_SAMP_SIZE_VAL)
> {
> +		wr_reg32(&r4tst->rtscmisc, (2 << 16) | 32);
> +		wr_reg32(&r4tst->rtpkrrng, 570);
> +		wr_reg32(&r4tst->rtpkrmax, 1600);
> +		wr_reg32(&r4tst->rtscml, (122 << 16) | 317);
> +		wr_reg32(&r4tst->rtscrl[0], (80 << 16) | 107);
> +		wr_reg32(&r4tst->rtscrl[1], (57 << 16) | 62);
> +		wr_reg32(&r4tst->rtscrl[2], (39 << 16) | 39);
> +		wr_reg32(&r4tst->rtscrl[3], (27 << 16) | 26);
> +		wr_reg32(&r4tst->rtscrl[4], (19 << 16) | 18);
> +		wr_reg32(&r4tst->rtscrl[5], (18 << 16) | 17);
> +	}
> +
>  	/*
>  	 * select raw sampling in both entropy shifter
>  	 * and statistical checker; ; put RNG4 into run mode diff --git
> a/drivers/crypto/caam/regs.h b/drivers/crypto/caam/regs.h index
> 66928f8a0c4b..189e74c21f0c 100644
> --- a/drivers/crypto/caam/regs.h
> +++ b/drivers/crypto/caam/regs.h
> @@ -3,7 +3,7 @@
>   * CAAM hardware register-level view
>   *
>   * Copyright 2008-2011 Freescale Semiconductor, Inc.
> - * Copyright 2018 NXP
> + * Copyright 2018, 2023 NXP
>   */
> 
>  #ifndef REGS_H
> @@ -523,6 +523,8 @@ struct rng4tst {
>  #define RTSDCTL_ENT_DLY_MASK (0xffff << RTSDCTL_ENT_DLY_SHIFT)
> #define RTSDCTL_ENT_DLY_MIN 3200  #define RTSDCTL_ENT_DLY_MAX 12800
> +#define RTSDCTL_SAMP_SIZE_MASK 0xffff
> +#define RTSDCTL_SAMP_SIZE_VAL 512
>  	u32 rtsdctl;		/* seed control register */
>  	union {
>  		u32 rtsblim;	/* PRGM=1: sparse bit limit register */
> @@ -534,7 +536,15 @@ struct rng4tst {
>  		u32 rtfrqmax;	/* PRGM=1: freq. count max. limit register */
>  		u32 rtfrqcnt;	/* PRGM=0: freq. count register */
>  	};
> -	u32 rsvd1[40];
> +	union {
> +		u32 rtscmc;	/* statistical check run monobit count */
> +		u32 rtscml;	/* statistical check run monobit limit */
> +	};
> +	union {
> +		u32 rtscrc[6];	/* statistical check run length count */
> +		u32 rtscrl[6];	/* statistical check run length limit */
> +	};
> +	u32 rsvd1[33];
>  #define RDSTA_SKVT 0x80000000
>  #define RDSTA_SKVN 0x40000000
>  #define RDSTA_PR0 BIT(4)
> --
> 2.25.1