On 11/5/21 10:34 PM, Julien Masson wrote: > With the recent changes on mmc driver, we saw that the boot is ~5 secs > longer compared to v2021.07 on mediatek platforms. > > This regression is seen during mmc_init and caused by the following > patch [1]. > > Indeed since we did not support poll dat0, we fulfilled the condition > of [1] and a delay of 500 ms was added for every __mmc_switch call. > > By adding the support of wait_dat0(), we now don't need to mdelay > during mmc_init anymore. > > [1]: > https://protect2.fireeye.com/v1/url?k=553fefc7-0aa4d6e8-553e6488-0cc47a31cdf8-658c3df5220dc893&q=1&e=badb6c30-0546-4072-80c5-a772143bc5ee&u=https%3A%2F%2Fpatchwork.ozlabs.org%2Fproject%2Fuboot%2Fpatch%2F1629192034-64056-1-git-send-email-ye.li%40nxp.com%2F > > Signed-off-by: Julien Masson <[email protected]>
Reviewed-by: Jaehoon Chung <[email protected]> Best Regards, Jaehoon Chung > --- > drivers/mmc/mtk-sd.c | 15 +++++++++++++++ > 1 file changed, 15 insertions(+) > > diff --git a/drivers/mmc/mtk-sd.c b/drivers/mmc/mtk-sd.c > index 8599f095bc..97182ffd7f 100644 > --- a/drivers/mmc/mtk-sd.c > +++ b/drivers/mmc/mtk-sd.c > @@ -1724,6 +1724,20 @@ static int msdc_drv_bind(struct udevice *dev) > return mmc_bind(dev, &plat->mmc, &plat->cfg); > } > > +static int msdc_ops_wait_dat0(struct udevice *dev, int state, int timeout_us) > +{ > + struct msdc_host *host = dev_get_priv(dev); > + int ret; > + u32 reg; > + > + ret = readl_poll_sleep_timeout(&host->base->msdc_ps, reg, > + !!(reg & MSDC_PS_DAT0) == !!state, > + 1000, /* 1 ms */ > + timeout_us); > + > + return ret; > +} > + > static const struct dm_mmc_ops msdc_ops = { > .send_cmd = msdc_ops_send_cmd, > .set_ios = msdc_ops_set_ios, > @@ -1732,6 +1746,7 @@ static const struct dm_mmc_ops msdc_ops = { > #ifdef MMC_SUPPORTS_TUNING > .execute_tuning = msdc_execute_tuning, > #endif > + .wait_dat0 = msdc_ops_wait_dat0, > }; > > static const struct msdc_compatible mt7620_compat = { >
