[v1,1/3] drivers: rng: add check status bit feature

For some Amlogic SOC's, mechanism to obtain random number
has been changed. For example, S4 now uses status bit waiting algo.

Signed-off-by: Alexey Romanov <avromanov@sberdevices.ru>
---
 drivers/char/hw_random/meson-rng.c | 77 ++++++++++++++++++++++++++++--
 1 file changed, 74 insertions(+), 3 deletions(-)

Hi Martin,

On Tue, Jul 25, 2023 at 09:59:01PM +0200, Martin Kaiser wrote:
> Alexey Romanov (avromanov@sberdevices.ru) wrote:
> 
> > For some Amlogic SOC's, mechanism to obtain random number
> > has been changed. For example, S4 now uses status bit waiting algo.
> 
> > Signed-off-by: Alexey Romanov <avromanov@sberdevices.ru>
> > ---
> >  drivers/char/hw_random/meson-rng.c | 77 ++++++++++++++++++++++++++++--
> >  1 file changed, 74 insertions(+), 3 deletions(-)
> 
> > diff --git a/drivers/char/hw_random/meson-rng.c b/drivers/char/hw_random/meson-rng.c
> > index a4eb8e35f13d..c6d7349630a1 100644
> > --- a/drivers/char/hw_random/meson-rng.c
> > +++ b/drivers/char/hw_random/meson-rng.c
> > @@ -14,19 +14,65 @@
> >  #include <linux/of.h>
> >  #include <linux/clk.h>
> 
> > -#define RNG_DATA 0x00
> > +struct meson_rng_priv {
> > +	bool check_status_bit;
> > +	unsigned int data_offset;
> > +	unsigned int cfg_offset;
> > +};
> 
> >  struct meson_rng_data {
> >  	void __iomem *base;
> >  	struct hwrng rng;
> > +	struct device *dev;
> > +	const struct meson_rng_priv *priv;
> >  };
> 
> > +#define RUN_BIT			0
> > +#define SEED_READY_STS_BIT	31
> > +#define RETRY_CNT		100
> > +
> > +static int meson_rng_wait_status(void __iomem *cfg_addr, int bit)
> > +{
> > +	u32 status;
> > +	u32 cnt = 0;
> > +
> > +	do {
> > +		status = readl_relaxed(cfg_addr) & BIT(bit);
> > +		cpu_relax();
> > +	} while (status && (cnt++ < RETRY_CNT));
> > +
> 
> Could you use readl_relaxed_poll_timeout here instead of open coding the
> loop?

At first I also thought about this API. But later I came to the
conclusion that it is inappropriate here:

1. We can't call rng_read from an atomic context.
2. RNG for me looks like a very lightweight primitive to me that 
should work quiclky.

But, now I looked again at the API and realized that we can use 
readl_relaxed_poll_timeout_atomic() instead of
readl_relaxed_poll_timeout(). What do you think?

> 
> > +	if (status)
> > +		return -EBUSY;
> > +
> > +	return 0;
> > +}
> > +
> >  static int meson_rng_read(struct hwrng *rng, void *buf, size_t max, bool wait)
> >  {
> >  	struct meson_rng_data *data =
> >  			container_of(rng, struct meson_rng_data, rng);
> > +	const struct meson_rng_priv *priv = data->priv;
> > +
> > +	if (priv->check_status_bit) {
> > +		void __iomem *cfg_addr = data->base + priv->cfg_offset;
> > +		int err;
> > +
> > +		writel_relaxed(readl_relaxed(cfg_addr) | BIT(SEED_READY_STS_BIT), cfg_addr);
> 
> > -	*(u32 *)buf = readl_relaxed(data->base + RNG_DATA);
> > +		err = meson_rng_wait_status(cfg_addr, SEED_READY_STS_BIT);
> > +		if (err) {
> > +			dev_err(data->dev, "Seed isn't ready, try again\n");
> > +			return err;
> > +		}
> > +
> > +		err = meson_rng_wait_status(cfg_addr, RUN_BIT);
> > +		if (err) {
> > +			dev_err(data->dev, "Can't get random number, try again\n");
> > +			return err;
> > +		}
> > +	}
> > +
> > +	*(u32 *)buf = readl_relaxed(data->base + priv->data_offset);
> 
> >  	return sizeof(u32);
> >  }
> > @@ -41,6 +87,10 @@ static int meson_rng_probe(struct platform_device *pdev)
> >  	if (!data)
> >  		return -ENOMEM;
> 
> > +	data->priv = device_get_match_data(&pdev->dev);
> > +	if (!data->priv)
> > +		return -ENODEV;
> > +
> >  	data->base = devm_platform_ioremap_resource(pdev, 0);
> >  	if (IS_ERR(data->base))
> >  		return PTR_ERR(data->base);
> > @@ -53,11 +103,32 @@ static int meson_rng_probe(struct platform_device *pdev)
> >  	data->rng.name = pdev->name;
> >  	data->rng.read = meson_rng_read;
> 
> > +	data->dev = &pdev->dev;
> > +
> >  	return devm_hwrng_register(dev, &data->rng);
> >  }
> 
> > +static const struct meson_rng_priv meson_rng_priv = {
> > +	.check_status_bit = false,
> > +	.data_offset = 0x0,
> > +	.cfg_offset = 0x0,
> > +};
> > +
> > +static const struct meson_rng_priv meson_rng_priv_s4 = {
> > +	.check_status_bit = true,
> > +	.data_offset = 0x8,
> > +	.cfg_offset = 0x0,
> > +};
> > +
> >  static const struct of_device_id meson_rng_of_match[] = {
> > -	{ .compatible = "amlogic,meson-rng", },
> > +	{
> > +		.compatible = "amlogic,meson-rng",
> > +		.data = (void *)&meson_rng_priv,
> > +	},
> > +	{
> > +		.compatible = "amlogic,meson-rng-s4",
> > +		.data = (void *)&meson_rng_priv_s4,
> > +	},
> >  	{},
> >  };
> >  MODULE_DEVICE_TABLE(of, meson_rng_of_match);
> > -- 
> > 2.38.1
>

Hi Alexey,

Alexey Romanov (AVRomanov@sberdevices.ru) wrote:

> > > +static int meson_rng_wait_status(void __iomem *cfg_addr, int bit)
> > > +{
> > > +	u32 status;
> > > +	u32 cnt = 0;
> > > +
> > > +	do {
> > > +		status = readl_relaxed(cfg_addr) & BIT(bit);
> > > +		cpu_relax();
> > > +	} while (status && (cnt++ < RETRY_CNT));
> > > +

> > Could you use readl_relaxed_poll_timeout here instead of open coding the
> > loop?

> At first I also thought about this API. But later I came to the
> conclusion that it is inappropriate here:

> 1. We can't call rng_read from an atomic context.

Agreed. A hwrng read function may sleep (an example for this is
exynos_trng_do_read). But this doesn't prevent us from using
readl_relaxed_poll_timeout.

> 2. RNG for me looks like a very lightweight primitive to me that 
> should work quiclky.

> But, now I looked again at the API and realized that we can use
> readl_relaxed_poll_timeout_atomic() instead of
> readl_relaxed_poll_timeout(). What do you think?

Ok, if you know that your rng hardware won't need much time to set the
bit that you're checking, you may use readl_relaxed_poll_timeout_atomic.

stm32_rtc_set_alarm does something similar.

Best regards,
Martin

Hello Alexey,

On Tue, Jul 25, 2023 at 4:13 PM Alexey Romanov <avromanov@sberdevices.ru> wrote:
[...]
>  static int meson_rng_read(struct hwrng *rng, void *buf, size_t max, bool wait)
>  {
>         struct meson_rng_data *data =
>                         container_of(rng, struct meson_rng_data, rng);
> +       const struct meson_rng_priv *priv = data->priv;
> +
> +       if (priv->check_status_bit) {
> +               void __iomem *cfg_addr = data->base + priv->cfg_offset;
> +               int err;
Have you considered just creating two separate functions:
- meson_rng_read
- meson_s4_rng_read

Then you don't need all of these if/else.
You can even skip the whole offset description in your new struct
meson_rng_priv.
All of this can make the code easier to understand.

> +               writel_relaxed(readl_relaxed(cfg_addr) | BIT(SEED_READY_STS_BIT), cfg_addr);
Why not just #define SEED_READY_STS as BIT(...) right away?


Best regards,
Martin