Received-SPF: pass (google.com: domain of linux-crypto-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Tue, 29 Aug 2023 20:58:13 +0200
From:   Martin Kaiser <martin@kaiser.cx>
To:     Alexander Stein <alexander.stein@ew.tq-group.com>
Cc:     Herbert Xu <herbert@gondor.apana.org.au>,
        linux-crypto@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH v2 3/6] hwrng: imx-rngc - use polling to detect end of
 self test
Message-ID: <20230829185813.odjl3zqwshbgki4m@viti.kaiser.cx>
References: <20230824192059.1569591-1-martin@kaiser.cx>
 <20230824192059.1569591-4-martin@kaiser.cx>
 <8370215.EvYhyI6sBW@steina-w>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <8370215.EvYhyI6sBW@steina-w>
User-Agent: NeoMutt/20170113 (1.7.2)
Sender: Martin Kaiser <martin@viti.kaiser.cx>
Precedence: bulk

Hi Alexander,

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

> I'm still not convinced that using polling is simpler. By using 
> readl_poll_timeout() you will also get an interrupt, the timer one. Why 
> exactly is using polling much (!) simpler?

it requires much less code in the driver.

> > The selftest should take approx. 450us. Keep the overhead to a minimum
> > by polling every 500us. (We've already lowered the timeout to 1.5ms.)

> I suppose these times only hold true for a specific peripheral clock 
> frequency. Is it guaranteed that this frequency is fixed?

Good point. The lowest possible peripheral clock frequency is half of
what I used for the calculations, i.e. 33.25MHz. That would double the
durations. Should be ok for the selftest. But for the initial seed, we'd
get into a region where readl_poll_timeout (usleep_range) does no longer
make sense.

> For using IRQ it's simpler, there is no guessing: you return once the self 
> test finished. The timeout is identical anyway.

I've looked at other callers of readl_poll_timeout. It seems that none
of them is called in a driver's probe function or uses an overall
timeout of 200ms.

I'll keep the interrupt + completion and resubmit the patches for
adjusting the timeouts.

Thanks,
Martin