Re: [PATCH] ath9k: use hw_random API instead of directly dumping into random.c

["Jason A. Donenfeld" <Jason@zx2c4.com>]

On 2/16/22, Toke Høiland-Jørgensen <toke@toke.dk> wrote:
> "Jason A. Donenfeld" <Jason@zx2c4.com> writes:
>
>> Hardware random number generators are supposed to use the hw_random
>> framework. This commit turns ath9k's kthread-based design into a proper
>> hw_random driver.
>>
>> This compiles, but I have no hardware or other ability to determine
>> whether it works. I'll leave further development up to the ath9k
>> and hw_random maintainers.
>>
>> Cc: Toke Høiland-Jørgensen <toke@redhat.com>
>> Cc: Kalle Valo <kvalo@kernel.org>
>> Cc: Dominik Brodowski <linux@dominikbrodowski.net>
>> Cc: Herbert Xu <herbert@gondor.apana.org.au>
>> Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
>> ---
>>  drivers/net/wireless/ath/ath9k/ath9k.h |  2 +-
>>  drivers/net/wireless/ath/ath9k/rng.c   | 62 +++++++++-----------------
>>  2 files changed, 23 insertions(+), 41 deletions(-)
>>
>> diff --git a/drivers/net/wireless/ath/ath9k/ath9k.h
>> b/drivers/net/wireless/ath/ath9k/ath9k.h
>> index ef6f5ea06c1f..142f472903dc 100644
>> --- a/drivers/net/wireless/ath/ath9k/ath9k.h
>> +++ b/drivers/net/wireless/ath/ath9k/ath9k.h
>> @@ -1072,7 +1072,7 @@ struct ath_softc {
>>
>>  #ifdef CONFIG_ATH9K_HWRNG
>>  	u32 rng_last;
>> -	struct task_struct *rng_task;
>> +	struct hwrng rng_ops;
>>  #endif
>>  };
>>
>> diff --git a/drivers/net/wireless/ath/ath9k/rng.c
>> b/drivers/net/wireless/ath/ath9k/rng.c
>> index aae2bd3cac69..369b222908ba 100644
>> --- a/drivers/net/wireless/ath/ath9k/rng.c
>> +++ b/drivers/net/wireless/ath/ath9k/rng.c
>> @@ -22,9 +22,6 @@
>>  #include "hw.h"
>>  #include "ar9003_phy.h"
>>
>> -#define ATH9K_RNG_BUF_SIZE	320
>> -#define ATH9K_RNG_ENTROPY(x)	(((x) * 8 * 10) >> 5) /* quality: 10/32 */
>
> So this comment says "quality: 10/32" but below you're setting "quality"
> as 320. No idea what the units are supposed to be, but is this right?

I think the unit is supposed to be how many entropic bits there are
out of 1024 bits? These types of estimates are always BS, so keeping
it on the lower end as before seemed right. Herbert can jump in here
if he has a better idea; that's his jig.

>
>>  static DECLARE_WAIT_QUEUE_HEAD(rng_queue);
>>
>>  static int ath9k_rng_data_read(struct ath_softc *sc, u32 *buf, u32
>> buf_size)
>
> This function takes buf as a *u32, and interprets buf_size as a number
> of u32s...

Oh my... Nice catch. I'll send a v2 shortly. I wonder how this managed
to work for Rui.

>
>> @@ -72,61 +69,46 @@ static u32 ath9k_rng_delay_get(u32 fail_stats)
>>  	return delay;
>>  }
>>
>> -static int ath9k_rng_kthread(void *data)
>> +static int ath9k_rng_read(struct hwrng *rng, void *buf, size_t max, bool
>> wait)
>>  {
>> +	struct ath_softc *sc = container_of(rng, struct ath_softc, rng_ops);
>>  	int bytes_read;
>> -	struct ath_softc *sc = data;
>> -	u32 *rng_buf;
>> -	u32 delay, fail_stats = 0;
>> -
>> -	rng_buf = kmalloc_array(ATH9K_RNG_BUF_SIZE, sizeof(u32), GFP_KERNEL);
>> -	if (!rng_buf)
>> -		goto out;
>> -
>> -	while (!kthread_should_stop()) {
>> -		bytes_read = ath9k_rng_data_read(sc, rng_buf,
>> -						 ATH9K_RNG_BUF_SIZE);
>> -		if (unlikely(!bytes_read)) {
>> -			delay = ath9k_rng_delay_get(++fail_stats);
>> -			wait_event_interruptible_timeout(rng_queue,
>> -							 kthread_should_stop(),
>> -							 msecs_to_jiffies(delay));
>> -			continue;
>> -		}
>> -
>> -		fail_stats = 0;
>> -
>> -		/* sleep until entropy bits under write_wakeup_threshold */
>> -		add_hwgenerator_randomness((void *)rng_buf, bytes_read,
>> -					   ATH9K_RNG_ENTROPY(bytes_read));
>> -	}
>> +	u32 fail_stats = 0;
>>
>> -	kfree(rng_buf);
>> -out:
>> -	sc->rng_task = NULL;
>> +retry:
>> +	bytes_read = ath9k_rng_data_read(sc, buf, max);
>
> ... but AFAICT here you're calling it with a buffer size from hw_random
> that's in bytes?

V2 on its way. Rui - you may need to re-test...

Jason