Date:   Wed, 30 Aug 2017 15:43:49 +0200
From:   Ulf Hansson <ulf.hansson@...aro.org>
To:     Hans de Goede <hdegoede@...hat.com>
Cc:     Quentin Schulz <quentin.schulz@...e-electrons.com>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Linus Walleij <linus.walleij@...aro.org>,
        Shawn Lin <shawn.lin@...k-chips.com>,
        Adrian Hunter <adrian.hunter@...el.com>,
        Baolin Wang <baolin.wang@...aro.org>,
        Maxime Ripard <maxime.ripard@...e-electrons.com>,
        Thomas Petazzoni <thomas.petazzoni@...e-electrons.com>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "linux-mmc@...r.kernel.org" <linux-mmc@...r.kernel.org>,
        driverdevel <devel@...verdev.osuosl.org>,
        Icenowy Zheng <icenowy@...c.xyz>, Chen-Yu Tsai <wens@...e.org>
Subject: Re: [PATCH 2/2] mmc: Add mmc_force_detect_change_begin / _end functions

On 30 August 2017 at 14:44, Hans de Goede <hdegoede@...hat.com> wrote:
> Hi,
>
>
> On 21-07-17 16:35, Quentin Schulz wrote:
>>
>> From: Hans de Goede <hdegoede@...hat.com>
>>
>> Some sdio devices have a multiple stage bring-up process. Specifically
>> the esp8089 (for which an out of tree driver is available) loads firmware
>> on the first call to its sdio-drivers' probe function and then resets
>> the device causing it to reboot from its RAM with the new firmware.
>>
>> When this sdio device reboots it comes back up in 1 bit 400 KHz mode
>> again, and we need to walk through the whole ios negatiation and sdio
>> setup
>> again.
>>
>> There are 2 problems with this:
>>
>> 1) Typically these devices are soldered onto some (ARM) tablet / SBC
>> PCB and as such are described in devicetree as "non-removable", which
>> causes the mmc-core to scan them only once and not poll for the device
>> dropping of the bus. Normally this is the right thing todo but in the
>> eso8089 example we need the mmc-core to notice the module has disconnected
>> (since it is now in 1 bit mode again it will not talk to the host in 4 bit
>> mode). This can be worked around by using "broken-cd" in devicetree
>> instead of "non-removable", but that is not a proper fix since the device
>> really is non-removable.
>>
>> 2) When the mmc-core detects the device has disconnected it will poweroff
>> the device, causing the RAM loaded firmware to be lost. This can be worked
>> around in devicetree by using regulator-always-on (and avoiding the use of
>> mmc-pwrseq), but again that is more of a hack then a proper fix.
>>
>> This commmit fixes 1) by adding a mmc_force_detect_change function which
>> will cause scanning for device removal / insertion until a new device is
>> detected. 2) Is fixed by a keep_power flag to the mmc_force_detect_change
>> function which when set causes the mmc-core to keep the power to the
>> device
>> on during the rescan.
>>
>> Cc: Icenowy Zheng <icenowy@...c.xyz>
>> Cc: Maxime Ripard <maxime.ripard@...e-electrons.com>
>> Cc: Chen-Yu Tsai <wens@...e.org>
>> Signed-off-by: Hans de Goede <hdegoede@...hat.com>
>
>
> So when I posted this patch quite a while back, there was some discussion
> about this and a consensus that this is not the right solution.
>
> So first of all lets describe the problem:
>
> The esp8089 sdio wifi chip is really an ARM core with a wifi phy
> connected to it (as many wifi chipsets are).
>
> But this one comes up in some really generic sdio capable boot-loader
> mode and we need to feed it firmware and then reboot it into the
> new firmware.
>
> The reboot is where the problems happens. It seems to fallback
> from the negotiated 4 wire sdio mode to single wire spi mode then.
>
> The out of tree version of the driver deals with this by not setting
> the non-removable flag as well as setting the broken_cd flag so that
> the mmc core polls the device, after the reboot the poll fails
> because the mmc-controller and the esp8089 are using a different
> amount of wires so the mmc-cmd the poll uses times out.
>
> After which the esp8089 drivers remove function gets called, and
> the mmc stack re-discovers the esp8089 by restarting the whole
> number of wires (and speed) used negotiation. After which the
> esp8089 driver's probe function gets called (again) and on
> firmware loading is has set a global flag, so now it actually
> acts as a wifi driver rather then trying to load the firmware
> a second time.
>
> Since I did not want to rely on broken_cd polling I came up
> with the hack which is this patch.
>
> So when this patch was first discussed we came to the conclusion
> that what we really need is some sort of mmc_reprobe_device
> function which the driver can call from probe which will
> redo the number of wires (and speed) used negotiation,
> while keeping the sdio_function device as is so that probe can
> simply continue after this and we also don't need the ugly
> global flag.
>
> The idea would be for this function to be some wrapper
> around mmc_init_card() which resets the ios settings as is
> normally done on remove and then call mmc_init_card()
> passing in the existing card the same way as is done
> one resume, so that the existing card / sdio_function
> devices get reused.
>
> IIRC Ulf would look into writing this mmc_reprobe_device
> function and then I would test it with the esp8089, but
> Ulf never got around to writing the function and I ended
> up working on other things too.

Thanks for summary!

Just to let you know, I haven't forgot about this problem. I am
planning for a major update of the SDIO for power management support,
within a not too far future.
The issue described above, is then also one of the things I also plan
to look into.

However, if there anybody that wants to hack on this, feel free to do it!

Kind regards
Uffe

Powered by blists - more mailing lists