On Thu, 24 Nov 2022 at 18:07, Adrian Hunter <adrian.hunter@xxxxxxxxx> wrote:
>
> Commit 20b92a30b561 ("mmc: sdhci: update signal voltage switch code")
> removed voltage switch delays from sdhci because mmc core had been
> enhanced to support them. However that assumed that sdhci_set_ios()
> did a single clock change, which it did not, and so the delays in mmc
> core, which should have come after the first clock change, were not
> effective.
>
> Fix by avoiding re-configuring UHS and preset settings when the clock
> is turning on and the settings have not changed. That then also avoids
> the associated clock changes, so that then sdhci_set_ios() does a single
> clock change when voltage switching, and the mmc core delays become
> effective.
>
> To do that has meant keeping track of driver strength (host->drv_type),
> and cases of reinitialization (host->reinit_uhs).
>
> Note also, the 'turning_on_clk' restriction should not be necessary
> but is done to minimize the impact of the change on stable kernels.
>
> Fixes: 20b92a30b561 ("mmc: sdhci: update signal voltage switch code")
> Cc: stable@xxxxxxxxxxxxxxx
> Signed-off-by: Adrian Hunter <adrian.hunter@xxxxxxxxx>
[...]
> @@ -2335,6 +2362,18 @@ void sdhci_set_ios(struct mmc_host *mmc, struct mmc_ios *ios)
>
> host->ops->set_bus_width(host, ios->bus_width);
>
> + /*
> + * Special case to avoid multiple clock changes during voltage
> + * switching.
> + */
> + if (!reinit_uhs &&
> + turning_on_clk &&
> + host->timing == ios->timing &&
> + host->version >= SDHCI_SPEC_300 &&
> + (host->preset_enabled || host->drv_type == ios->drv_type) &&
> + (host->preset_enabled || !sdhci_preset_needed(host, ios->timing)))
This caught my eyes, as it looks a bit messy. Not sure if it's
possible to simplify, but the last two lines could be replaced with:
(host->preset_enabled ||
(host->drv_type == ios->drv_type && !sdhci_preset_needed(host, ios->timing))))
But, I am not sure that really helps...
[...]
Kind regards
Uffe
