mbox series

[0/3] hwrng: imx-rngc - use polling instead of interrupt

Message ID 20230822152553.190858-1-martin@kaiser.cx
Headers show
Series hwrng: imx-rngc - use polling instead of interrupt | expand

Message

Martin Kaiser Aug. 22, 2023, 3:25 p.m. UTC
Use polling and wait actively for the self test and the initial seeding
of the rngc to complete. This is much simpler than using an interrupt
and a completion. Also, there's nothing we can do in parallel while self
test or initial seeding are running.

Martin Kaiser (3):
  hwrng: imx-rngc - use polling to detect end of self test
  hwrng: imx-rngc - read status register for error checks
  hwrng: imx-rngc - use polling to wait for end of seeding

 drivers/char/hw_random/imx-rngc.c | 96 ++++++-------------------------
 1 file changed, 16 insertions(+), 80 deletions(-)

Comments

Alexander Stein Aug. 23, 2023, 5:23 a.m. UTC | #1
Am Dienstag, 22. August 2023, 17:25:51 CEST schrieb Martin Kaiser:
> Use polling to detect the end of the rngc self test. This is much simpler
> than using an interrupt and a completion.
> 
> Active waiting is no disadvantage here. The self test is run during
> probe, there's nothing we could do in parallel at this time.

If this driver is built-in you are stalling the boot process while polling, 
no? Unless probe_type = PROBE_PREFER_ASYNCHRONOUS is set of course.

Best regards,
Alexander

> Signed-off-by: Martin Kaiser <martin@kaiser.cx>
> ---
>  drivers/char/hw_random/imx-rngc.c | 15 +++++++--------
>  1 file changed, 7 insertions(+), 8 deletions(-)
> 
> diff --git a/drivers/char/hw_random/imx-rngc.c
> b/drivers/char/hw_random/imx-rngc.c index e4b385b01b11..85207535fd12 100644
> --- a/drivers/char/hw_random/imx-rngc.c
> +++ b/drivers/char/hw_random/imx-rngc.c
> @@ -17,6 +17,7 @@
>  #include <linux/hw_random.h>
>  #include <linux/completion.h>
>  #include <linux/io.h>
> +#include <linux/iopoll.h>
>  #include <linux/bitfield.h>
> 
>  #define RNGC_VER_ID			0x0000
> @@ -101,21 +102,19 @@ static inline void imx_rngc_irq_unmask(struct imx_rngc
> *rngc)
> 
>  static int imx_rngc_self_test(struct imx_rngc *rngc)
>  {
> -	u32 cmd;
> +	u32 cmd, status;
>  	int ret;
> 
> -	imx_rngc_irq_unmask(rngc);
> -
>  	/* run self test */
>  	cmd = readl(rngc->base + RNGC_COMMAND);
>  	writel(cmd | RNGC_CMD_SELF_TEST, rngc->base + RNGC_COMMAND);
> 
> -	ret = wait_for_completion_timeout(&rngc->rng_op_done,
> msecs_to_jiffies(RNGC_TIMEOUT)); -	imx_rngc_irq_mask_clear(rngc);
> -	if (!ret)
> -		return -ETIMEDOUT;
> +	ret = readl_poll_timeout(rngc->base + RNGC_STATUS, status,
> +				 status & RNGC_STATUS_ST_DONE, 1000, 
RNGC_TIMEOUT * 1000);
> +	if (ret < 0)
> +		return ret;
> 
> -	return rngc->err_reg ? -EIO : 0;
> +	return readl(rngc->base + RNGC_ERROR) ? -EIO : 0;
>  }
> 
>  static int imx_rngc_read(struct hwrng *rng, void *data, size_t max, bool
> wait)
Alexander Stein Aug. 23, 2023, 5:33 a.m. UTC | #2
Am Dienstag, 22. August 2023, 17:25:53 CEST schrieb Martin Kaiser:
> Use polling to wait until the imx-rngc is properly seeded.

What is the benefit of burning CPU cycles while waiting for hardware?

> We do this only in the init function when the imx-rngc becomes active.
> Polling is ok at this time, there's nothing else we could do while
> we're waiting.
> 
> We can now remove the code for the interrupt and the completion.

Please split the change to polling and IRQ removal into two patches.

> Signed-off-by: Martin Kaiser <martin@kaiser.cx>
> ---
>  drivers/char/hw_random/imx-rngc.c | 81 ++++---------------------------
>  1 file changed, 9 insertions(+), 72 deletions(-)
> 
> diff --git a/drivers/char/hw_random/imx-rngc.c
> b/drivers/char/hw_random/imx-rngc.c index d2df468fd460..7ab9aada72d0 100644
> --- a/drivers/char/hw_random/imx-rngc.c
> +++ b/drivers/char/hw_random/imx-rngc.c
> @@ -63,12 +63,6 @@ struct imx_rngc {
>  	struct clk		*clk;
>  	void __iomem		*base;
>  	struct hwrng		rng;
> -	struct completion	rng_op_done;
> -	/*
> -	 * err_reg is written only by the irq handler and read only
> -	 * when interrupts are masked, we need no spinlock
> -	 */
> -	u32			err_reg;
>  };
> 
> 
> @@ -91,15 +85,6 @@ static inline void imx_rngc_irq_mask_clear(struct
> imx_rngc *rngc) writel(cmd, rngc->base + RNGC_COMMAND);
>  }
> 
> -static inline void imx_rngc_irq_unmask(struct imx_rngc *rngc)
> -{
> -	u32 ctrl;
> -
> -	ctrl = readl(rngc->base + RNGC_CONTROL);
> -	ctrl &= ~(RNGC_CTRL_MASK_DONE | RNGC_CTRL_MASK_ERROR);
> -	writel(ctrl, rngc->base + RNGC_CONTROL);
> -}
> -
>  static int imx_rngc_self_test(struct imx_rngc *rngc)
>  {
>  	u32 cmd, status;
> @@ -143,56 +128,32 @@ static int imx_rngc_read(struct hwrng *rng, void
> *data, size_t max, bool wait) return retval ? retval : -EIO;
>  }
> 
> -static irqreturn_t imx_rngc_irq(int irq, void *priv)
> -{
> -	struct imx_rngc *rngc = (struct imx_rngc *)priv;
> -	u32 status;
> -
> -	/*
> -	 * clearing the interrupt will also clear the error register
> -	 * read error and status before clearing
> -	 */
> -	status = readl(rngc->base + RNGC_STATUS);
> -	rngc->err_reg = readl(rngc->base + RNGC_ERROR);
> -
> -	imx_rngc_irq_mask_clear(rngc);
> -
> -	if (status & (RNGC_STATUS_SEED_DONE | RNGC_STATUS_ST_DONE))
> -		complete(&rngc->rng_op_done);
> -
> -	return IRQ_HANDLED;
> -}
> -
>  static int imx_rngc_init(struct hwrng *rng)
>  {
>  	struct imx_rngc *rngc = container_of(rng, struct imx_rngc, rng);
> -	u32 cmd, ctrl;
> +	u32 cmd, ctrl, status, err_reg;
>  	int ret;
> 
>  	/* clear error */
>  	cmd = readl(rngc->base + RNGC_COMMAND);
>  	writel(cmd | RNGC_CMD_CLR_ERR, rngc->base + RNGC_COMMAND);
> 
> -	imx_rngc_irq_unmask(rngc);
> -
>  	/* create seed, repeat while there is some statistical error */
>  	do {
>  		/* seed creation */
>  		cmd = readl(rngc->base + RNGC_COMMAND);
>  		writel(cmd | RNGC_CMD_SEED, rngc->base + RNGC_COMMAND);
> 
> -		ret = wait_for_completion_timeout(&rngc->rng_op_done,
> msecs_to_jiffies(RNGC_TIMEOUT)); -		if (!ret) {
> -			ret = -ETIMEDOUT;
> -			goto err;
> -		}
> +		ret = readl_poll_timeout(rngc->base + RNGC_STATUS, status,
> +					 status & 
RNGC_STATUS_SEED_DONE, 1000, RNGC_TIMEOUT * 1000);

So you want to poll for up to 3s?

Best regards,
Alexander

> +		if (ret < 0)
> +			return ret;
> 
> -	} while (rngc->err_reg == RNGC_ERROR_STATUS_STAT_ERR);
> +		err_reg = readl(rngc->base + RNGC_ERROR);
> +	} while (err_reg == RNGC_ERROR_STATUS_STAT_ERR);
> 
> -	if (rngc->err_reg) {
> -		ret = -EIO;
> -		goto err;
> -	}
> +	if (err_reg)
> +		return -EIO;
> 
>  	/*
>  	 * enable automatic seeding, the rngc creates a new seed 
automatically
> @@ -202,23 +163,7 @@ static int imx_rngc_init(struct hwrng *rng)
>  	ctrl |= RNGC_CTRL_AUTO_SEED;
>  	writel(ctrl, rngc->base + RNGC_CONTROL);
> 
> -	/*
> -	 * if initialisation was successful, we keep the interrupt
> -	 * unmasked until imx_rngc_cleanup is called
> -	 * we mask the interrupt ourselves if we return an error
> -	 */
>  	return 0;
> -
> -err:
> -	imx_rngc_irq_mask_clear(rngc);
> -	return ret;
> -}
> -
> -static void imx_rngc_cleanup(struct hwrng *rng)
> -{
> -	struct imx_rngc *rngc = container_of(rng, struct imx_rngc, rng);
> -
> -	imx_rngc_irq_mask_clear(rngc);
>  }
> 
>  static int __init imx_rngc_probe(struct platform_device *pdev)
> @@ -254,12 +199,9 @@ static int __init imx_rngc_probe(struct platform_device
> *pdev) if (rng_type != RNGC_TYPE_RNGC && rng_type != RNGC_TYPE_RNGB)
>  		return -ENODEV;
> 
> -	init_completion(&rngc->rng_op_done);
> -
>  	rngc->rng.name = pdev->name;
>  	rngc->rng.init = imx_rngc_init;
>  	rngc->rng.read = imx_rngc_read;
> -	rngc->rng.cleanup = imx_rngc_cleanup;
>  	rngc->rng.quality = 19;
> 
>  	rngc->dev = &pdev->dev;
> @@ -267,11 +209,6 @@ static int __init imx_rngc_probe(struct platform_device
> *pdev)
> 
>  	imx_rngc_irq_mask_clear(rngc);
> 
> -	ret = devm_request_irq(&pdev->dev,
> -			irq, imx_rngc_irq, 0, pdev->name, (void *)rngc);
> -	if (ret)
> -		return dev_err_probe(&pdev->dev, ret, "Can't get interrupt 
working.\n");
> -
>  	if (self_test) {
>  		ret = imx_rngc_self_test(rngc);
>  		if (ret)
Martin Kaiser Aug. 23, 2023, 11:07 a.m. UTC | #3
Hi Alexander,

thanks for reviewing the patches.

Alexander Stein (alexander.stein@ew.tq-group.com) wrote:

> Am Dienstag, 22. August 2023, 17:25:51 CEST schrieb Martin Kaiser:
> > Use polling to detect the end of the rngc self test. This is much simpler
> > than using an interrupt and a completion.

> > Active waiting is no disadvantage here. The self test is run during
> > probe, there's nothing we could do in parallel at this time.

> If this driver is built-in you are stalling the boot process while
> polling, no?

According to the manual, "The self test takes approximately 29,000
cycles to complete." On my system, the rngb peripheral clock runs at
66.5MHz, i.e. the self test takes 436us.

A poll interval of 480us and a timeout of 1.5ms are more appropriate
than the current values (1ms poll interval, 3sec timeout). I'll fix this
in v2.

If it is unacceptable for a probe function to run for 1.5ms, we could
probably move the self test to the init function. It's called when the
rng is selected as the active rng.

> Unless probe_type = PROBE_PREFER_ASYNCHRONOUS is set of
> course.

This does not work for drivers that use module_platform_driver_probe.

We could switch to module_platform_driver. I'd prefer fixing the timing
or moving the self test to init, though.

Best regards,
Martin
Martin Kaiser Aug. 23, 2023, 11:16 a.m. UTC | #4
Hi Alexander,

Alexander Stein (alexander.stein@ew.tq-group.com) wrote:

> Am Dienstag, 22. August 2023, 17:25:53 CEST schrieb Martin Kaiser:
> > Use polling to wait until the imx-rngc is properly seeded.

> What is the benefit of burning CPU cycles while waiting for hardware?

it seems to me that

readl_poll_timeout
   readx_poll_timeout
      usleep_range
         usleep_range_state
            schedule_hrtimeout_range

will not wait in the foreground and burn CPU cycles.

The comment for schedule_hrtimeout_range says

 * Make the current task sleep until the given expiry time has
 * elapsed.

> > We do this only in the init function when the imx-rngc becomes active.
> > Polling is ok at this time, there's nothing else we could do while
> > we're waiting.

> > We can now remove the code for the interrupt and the completion.

> Please split the change to polling and IRQ removal into two patches.

Good point. Will do this in v2.

> RNGC_STATUS_SEED_DONE, 1000, RNGC_TIMEOUT * 1000);
> So you want to poll for up to 3s?

According to the manual, "The initial seed takes approximately 2,000,000
clock cycles." With a 66.5MHz clock, this is approx 30ms. So that should
be

RNGC_STATUS_SEED_DONE, 20000, 100000);

The comment for readx_poll_timeout suggests <= 20ms for the poll
interval.

Best regards,
Martin