Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <a40dbff1-6aa1-dedf-7eb2-c75d75d808b2@marcan.st>
Date:   Wed, 2 Nov 2022 03:28:43 +0900
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.9.1
Subject: Re: [PATCH v3] i2c/pasemi: PASemi I2C controller IRQ enablement
Content-Language: en-US
To:     Arminder Singh <arminders208@outlook.com>,
        linux-kernel@vger.kernel.org
Cc:     linux-i2c@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
        asahi@lists.linux.dev, linuxppc-dev@lists.ozlabs.org,
        Alyssa Rosenzweig <alyssa@rosenzweig.io>,
        Sven Peter <sven@svenpeter.dev>,
        Christophe Leroy <christophe.leroy@csgroup.eu>,
        Nicholas Piggin <npiggin@gmail.com>,
        Michael Ellerman <mpe@ellerman.id.au>
References: <MN2PR01MB5358ED8FC32C0CFAEBD4A0E19F5F9@MN2PR01MB5358.prod.exchangelabs.com>
From:   Hector Martin <marcan@marcan.st>
In-Reply-To: <MN2PR01MB5358ED8FC32C0CFAEBD4A0E19F5F9@MN2PR01MB5358.prod.exchangelabs.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 07/10/2022 09.42, Arminder Singh wrote:
> This patch adds IRQ support to the PASemi I2C controller driver to 
> increase the performace of I2C transactions on platforms with PASemi I2C 
> controllers. While primarily intended for Apple silicon platforms, this 
> patch should also help in enabling IRQ support for older PASemi hardware 
> as well should the need arise.
> 
> Signed-off-by: Arminder Singh <arminders208@outlook.com>
> ---
> This version of the patch has been tested on an M1 Ultra Mac Studio,
> as well as an M1 MacBook Pro, and userspace launches successfully
> while using the IRQ path for I2C transactions.
>
[...]

Please increase the timeout to 100ms for v4. 10ms was always wrong (the
datasheet says the hardware clock stretching timeout is 25ms, and most
i2c drivers have much larger timeouts), and with the tighter timing
achievable with the IRQ patchset we are seeing timeouts in tipd
controller requests which can clock-stretch for ~10ms themselves,
followed by a spiral of errors as the driver has pretty poor error
recovery. Increasing the timeout fixes the immediate problem/regression.

Other than that, I now have a patch that makes the whole timeout/error
detection/recovery much more robust, but I can submit it after this goes
in :)

- Hector