Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20210204120547.15381-1-yann.gautier@foss.st.com> <20210204120547.15381-2-yann.gautier@foss.st.com>
In-Reply-To: <20210204120547.15381-2-yann.gautier@foss.st.com>
From:   Ulf Hansson <ulf.hansson@linaro.org>
Date:   Fri, 5 Feb 2021 10:53:34 +0100
Message-ID: <CAPDyKFqdtK33HSW_AM0s9172V=cBM6wnKuHubXSOGCVqJ8nzFg@mail.gmail.com>
Subject: Re: [PATCH 1/2] mmc: mmci: enable MMC_CAP_NEED_RSP_BUSY
To:     Yann Gautier <yann.gautier@foss.st.com>
Cc:     Russell King <linux@armlinux.org.uk>,
        Linus Walleij <linus.walleij@linaro.org>,
        ludovic.barre@foss.st.com,
        =?UTF-8?B?TWFyZWsgVmHFoXV0?= <marex@denx.de>,
        "linux-mmc@vger.kernel.org" <linux-mmc@vger.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

- trimmed cc-list

On Thu, 4 Feb 2021 at 13:08, <yann.gautier@foss.st.com> wrote:
>
> From: Yann Gautier <yann.gautier@foss.st.com>
>
> To properly manage commands awaiting R1B responses, the capability
> MMC_CAP_NEED_RSP_BUSY is enabled in mmci driver, for variants that
> manage busy detection.
> This R1B management needs both the flags MMC_CAP_NEED_RSP_BUSY and
> MMC_CAP_WAIT_WHILE_BUSY to be enabled together.

Would it be possible for you to share a little bit more about the
problem? Like under what circumstances does things screw up?

Is the issue only occurring when the cmd->busy_timeout becomes larger
than host->max_busy_timeout. Or even in other cases?

>
> Signed-off-by: Yann Gautier <yann.gautier@foss.st.com>
> ---
>  drivers/mmc/host/mmci.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/mmc/host/mmci.c b/drivers/mmc/host/mmci.c
> index 1bc674577ff9..bf6971fdd1a6 100644
> --- a/drivers/mmc/host/mmci.c
> +++ b/drivers/mmc/host/mmci.c
> @@ -2148,7 +2148,7 @@ static int mmci_probe(struct amba_device *dev,
>                 if (variant->busy_dpsm_flag)
>                         mmci_write_datactrlreg(host,
>                                                host->variant->busy_dpsm_flag);
> -               mmc->caps |= MMC_CAP_WAIT_WHILE_BUSY;
> +               mmc->caps |= MMC_CAP_WAIT_WHILE_BUSY | MMC_CAP_NEED_RSP_BUSY;

This isn't correct as the ux500 (and likely also other legacy
variants) don't need this. I have tried it in the past and it works
fine for ux500 without MMC_CAP_NEED_RSP_BUSY.

The difference is rather that the busy detection for stm32 variants
needs a corresponding HW busy timeout to be set (its
variant->busy_timeout flag is set). Perhaps we can use that
information instead?

Note that, MMC_CAP_NEED_RSP_BUSY, means that cmd->busy_timeout will
not be set by the core for erase commands, CMD5 and CMD6.

By looking at the code in mmci_start_command(), it looks like we will
default to a timeout of 10s, when cmd->busy_timeout isn't set. At
least for some erase requests, that won't be sufficient. Would it be
possible to disable the HW busy timeout in some way - and maybe use a
software timeout instead? Maybe I already asked Ludovic about this?
:-)

BTW, did you check that the MMCIDATATIMER does get the correct value
set for the timer in mmci_start_command() and if
host->max_busy_timeout gets correctly set in
mmci_set_max_busy_timeout()?

[...]

Kind regards
Uffe