On 13/03/2023 14:55, Marc Gonzalez wrote:
> Still trying to bisect this heisenbug into submission... :(
>
> So far, I've pared it down to mmc_attach_sdio()
>
> When probe WORKS, mmc_attach_sdio() returns 0.
> When probe FAILS, mmc_attach_sdio() returns ETIMEDOUT
> via mmc_send_io_op_cond(host, 0, &ocr);
>
> Wrapping mmc_send_io_op_cond() in a loop
> makes it work on the second try.
Almost there, I think, I hope :)
DT prop "post-power-on-delay-ms" looks like what I needed all along.
It exists both for host (default 10 ms) and for pwrseq_simple (default 0 apparently).
/*
* Apply power to the MMC stack. This is a two-stage process.
* First, we enable power to the card without the clock running.
* We then wait a bit for the power to stabilise. Finally,
* enable the bus drivers and clock to the card.
*
* We must _NOT_ enable the clock prior to power stablising.
*
* If a host does all the power sequencing itself, ignore the
* initial MMC_POWER_UP stage.
*/
void mmc_power_up(struct mmc_host *host, u32 ocr)
Calls:
mmc_delay(host->ios.power_delay_ms);
mmc_pwrseq_post_power_on(host);
=> msleep(pwrseq->post_power_on_delay_ms);
...
mmc_delay(host->ios.power_delay_ms);
QUESTION:
It's not clear to me why we sleep twice for host->ios.power_delay_ms?
Looks like all I need is to add post-power-on-delay-ms = <100>; to
sdio_pwrseq: sdio-pwrseq {
compatible = "mmc-pwrseq-simple";
reset-gpios = <&gpio GPIOX_6 GPIO_ACTIVE_LOW>;
clocks = <&wifi32k>;
clock-names = "ext_clock";
};
Will revert all my stabs in the dark, and boot the board 100 times
to check if this does the trick.
Regards
