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.

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)
-- 
2.39.2

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)


-- 
TQ-Systems GmbH | Mühlstraße 2, Gut Delling | 82229 Seefeld, Germany
Amtsgericht München, HRB 105018
Geschäftsführer: Detlef Schneider, Rüdiger Stahl, Stefan Schneider
http://www.tq-group.com/

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