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: <c5252887-e4f9-953d-88df-d8cc62ecd136@eaxlabs.cz>
In-Reply-To: <c5252887-e4f9-953d-88df-d8cc62ecd136@eaxlabs.cz>
From:   Ulf Hansson <ulf.hansson@linaro.org>
Date:   Mon, 30 Nov 2020 16:08:03 +0100
Message-ID: <CAPDyKFr8t-u_4YkuCzQ3Xd3VgxkTq_Anu+NzBo4Pfj8SLBnsvg@mail.gmail.com>
Subject: Re: armmmci rmmod causes hung tasks
To:     Martin DEVERA <devik@eaxlabs.cz>
Cc:     Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Linux ARM <linux-arm-kernel@lists.infradead.org>,
        "linux-mmc@vger.kernel.org" <linux-mmc@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Sun, 29 Nov 2020 at 19:20, Martin DEVERA <devik@eaxlabs.cz> wrote:
>
> Hello,
>
> on STM32MP1 with almost vanilla 5.7.7 in single CPU mode. Pair of
> modprobe armmmci ; rmmod armmmci
>
> causes rmmod and kworker to hang. I should note that no MMC is detected
> on the board (SDIO device on MMC bus is not responding).
> On another board (where SDIO is responding) rmmod works.
>
> It seems as another manifestation of https://lkml.org/lkml/2019/8/27/945
>
> Thanks.
>
> INFO: task kworker/0:1:12 blocked for more than 368 seconds.
>        Not tainted 5.7.7kdb-00003-g10397828596c-dirty #224
> "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
> kworker/0:1     D    0    12      2 0x00000000
> Workqueue: events_freezable mmc_rescan
> (__schedule) from (schedule+0x5b/0x90)
> (schedule) from (schedule_timeout+0x1b/0xa0)
> (schedule_timeout) from (__wait_for_common+0x7d/0xdc)
> (__wait_for_common) from (mmc_wait_for_req_done+0x1b/0x8c)
> (mmc_wait_for_req_done) from (mmc_wait_for_cmd+0x4d/0x68)
> (mmc_wait_for_cmd) from (mmc_io_rw_direct_host+0x87/0xc8)
> (mmc_io_rw_direct_host) from (sdio_reset+0x3b/0x58)
> (sdio_reset) from (mmc_rescan+0x15d/0x1d4)
> (mmc_rescan) from (process_one_work+0xdd/0x168)
> (process_one_work) from (worker_thread+0x17d/0x1ec)
> (worker_thread) from (kthread+0x9b/0xa4)
> (kthread) from (ret_from_fork+0x11/0x28)

It looks like the worker thread, which runs mmc_rescan() to try to
detect the SDIO card is hanging. Exactly why, I don't know.

Could be a misconfigured clock, pinctrl or a power domain being
suddenly gated...

>
> INFO: task rmmod:308 blocked for more than 368 seconds.
>        Not tainted 5.7.7kdb-00003-g10397828596c-dirty #224
> "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
> rmmod           D    0   308    222 0x00000000
> (__schedule) from (schedule+0x5b/0x90)
> (schedule) from (schedule_timeout+0x1b/0xa0)
> (schedule_timeout) from (__wait_for_common+0x7d/0xdc)
> (__wait_for_common) from (__flush_work+0xb3/0x110)
> (__flush_work) from (__cancel_work_timer+0x97/0xf0)
> (__cancel_work_timer) from (mmc_stop_host+0x29/0x88)
> (mmc_stop_host) from (mmc_remove_host+0x9/0x22)
> (mmc_remove_host) from (mmci_remove+0x19/0x92 [armmmci])
> (mmci_remove [armmmci]) from (amba_remove+0x13/0x42)
> (amba_remove) from (device_release_driver_internal+0x89/0xe4)
> (device_release_driver_internal) from (driver_detach+0x57/0x60)
> (driver_detach) from (bus_remove_driver+0x53/0x84)
> (bus_remove_driver) from (mmci_driver_exit+0x9/0x11dc [armmmci])
> (mmci_driver_exit [armmmci]) from (sys_delete_module+0x115/0x154)
> (sys_delete_module) from (ret_fast_syscall+0x1/0x5a)
>

When unbinding the mmc host, mmc_stop_host() needs to make sure that
mmc_rescan() becomes disabled. This is to prevent a card from being
detected/removed at same time as when the host is being removed.

In any case, when we call...

cancel_delayed_work_sync(&host->detect);

...it seems like we end up waiting for the currently running
mmc_rescan() work to be completed.

Kind regards
Uffe