Re: [PATCH v5 08/24] wfx: add bus_sdio.c

[Ulf Hansson <ulf.hansson@linaro.org>]

On Fri, 1 Oct 2021 at 14:31, Kalle Valo <kvalo@codeaurora.org> wrote:
>
> Hi Ulf,
>
> sorry for the late reply, my Gnus tells me it took me 24 weeks to reply :)
>
> Ulf Hansson <ulf.hansson@linaro.org> writes:
>
> > On Wed, 7 Apr 2021 at 14:00, Kalle Valo <kvalo@codeaurora.org> wrote:
> >>
> >> Ulf Hansson <ulf.hansson@linaro.org> writes:
> >>
> >> >> If I follow what has been done in other drivers I would write something
> >> >> like:
> >> >>
> >> >>   static int wfx_sdio_suspend(struct device *dev)
> >> >>   {
> >> >>           struct sdio_func *func = dev_to_sdio_func(dev);
> >> >>           struct wfx_sdio_priv *bus = sdio_get_drvdata(func);
> >> >>
> >> >>           config_reg_write_bits(bus->core, CFG_IRQ_ENABLE_DATA, 0);
> >> >>           // Necessary to keep device firmware in RAM
> >> >>           return sdio_set_host_pm_flags(func, MMC_PM_KEEP_POWER);
> >> >
> >> > This will tell the mmc/sdio core to keep the SDIO card powered on
> >> > during system suspend. Thus, it doesn't need to re-initialize it at
> >> > system resume - and the firmware should not need to be re-programmed.
> >> >
> >> > On the other hand, if you don't plan to support system wakeups, it
> >> > would probably be better to power off the card, to avoid wasting
> >> > energy while the system is suspended. I assume that means you need to
> >> > re-program the firmware as well. Normally, it's these kinds of things
> >> > that need to be managed from a ->resume() callback.
> >>
> >> Many mac80211 drivers do so that the device is powered off during
> >> interface down (ifconfig wlan0 down), and as mac80211 does interface
> >> down automatically during suspend, suspend then works without extra
> >> handlers.
> >
> > That sounds simple. :-)
>
> Indeed, I was omitting a lot of details :) My comment was more like a
> general remark to all different bus techonologies, not just about SDIO.
> And I'm not saying that all wireless drivers do that, but some of them
> do. Though I don't have any numbers how many.
>
> > Would you mind elaborating on what is actually being powered off at
> > interface down - and thus also I am curious what happens at a typical
> > interface up?
>
> In general in the drivers that do we this the firmware is completely
> turned off and all memory is reset during interface down. And firmware
> is started from the scratch during interface up. Also one benefit from
> this is that firmware state is reset, the wireless firmwares are
> notarious being buggy.
>
> > Even if we don't want to use system wakeups (wake-on-lan), the SDIO
> > core and the SDIO func driver still need to somewhat agree on how to
> > manage the power for the card during system suspend, I think.
> >
> > For example, for a non-removable SDIO card, the SDIO/MMC core may
> > decide to power off the card in system suspend. Then it needs to
> > restore power to the card and re-initialize it at system resume, of
> > course. This doesn't mean that the actual corresponding struct device
> > for it, gets removed/re-added, thus the SDIO func driver isn't being
> > re-probed after the system has resumed. Although, since the SDIO card
> > was re-initialized, it's likely that the FW may need to be
> > re-programmed after the system has been resumed.
> >
> > Are you saying that re-programming the FW is always happening at
> > interface up, when there are none system suspend/resume callbacks
> > assigned for the SDIO func driver?
>
> Yes, that's what I was trying to say. But take all this with grain of
> salt, I'm not very familiar with SDIO! And funnily enough, I checked
> what we do in ath10k_sdio driver during suspend has conflicting code and
> documentation:
>
> /* Empty handlers so that mmc subsystem doesn't remove us entirely during
>  * suspend. We instead follow cfg80211 suspend/resume handlers.
>  */
> static int ath10k_sdio_pm_suspend(struct device *device)
> {
>         struct sdio_func *func = dev_to_sdio_func(device);
>         struct ath10k_sdio *ar_sdio = sdio_get_drvdata(func);
>         struct ath10k *ar = ar_sdio->ar;
>         mmc_pm_flag_t pm_flag, pm_caps;
>         int ret;
>
>         if (!device_may_wakeup(ar->dev))
>                 return 0;
>
>         ath10k_sdio_set_mbox_sleep(ar, true);
>
>         pm_flag = MMC_PM_KEEP_POWER;
>
>         ret = sdio_set_host_pm_flags(func, pm_flag);
>         if (ret) {
>                 pm_caps = sdio_get_host_pm_caps(func);
>                 ath10k_warn(ar, "failed to set sdio host pm flags (0x%x, 0x%x): %d\n",
>                             pm_flag, pm_caps, ret);
>                 return ret;
>         }
>
>         return ret;

Just to confirm, the code looks reasonable to me, even if the comment
above looks a bit odd/outdated. :-)

*) Because the SDIO driver's ->suspend|resume() callbacks have been
assigned, the mmc core will not remove the corresponding SDIO
func/card's struct device.

**) If system wakeup *isn't* going to be enabled, the early return
with 0, will allow the mmc core to power off the SDIO card/func device
during system suspend. Vice versa, it will then restore power to it
and re-initialize it during system resume.

***) If system wakeup *is* going to be enabled, MMC_PM_KEEP_POWER flag
will prevent the mmc core from powering off the SDIO card/func device
during system suspend. Depending on if the wakeup irq is in-band or
out-band, MMC_PM_WAKE_SDIO_IRQ could be set too.

That said, note that ->probe() of the SDIO func driver, will not be
called for a non-removable SDIO func/card to re-program the FW after a
system suspend/resume. That needs to be managed from the SDIO func
driver's system resume callback - or deferring to upper common network
layers (interface up?), which seems to be the case here.

Kind regards
Uffe