| Message ID | 1515581362-181200-1-git-send-email-shawn.lin@rock-chips.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi, This seems like a good idea to me. The fact that there was no polling loop here always seemed strange to me, but that's how the original code was structured and I personally never saw any problems with it. On Wed, Jan 10, 2018 at 2:49 AM, Shawn Lin <shawn.lin@rock-chips.com> wrote: > It turns out that 5us isn't enough for all cases, so let's > retry some more times to wait for caldone. > > Signed-off-by: Shawn Lin <shawn.lin@rock-chips.com> > Reviewed-by: Brian Norris <briannorris@chromium.org> > Tested-by: Caesar Wang <wxt@rock-chips.com> > Tested-by: Ziyuan Xu <xzy.xu@rock-chips.com> > --- > > Changes in v2: > - propagate the error and print it > > drivers/phy/rockchip/phy-rockchip-emmc.c | 27 +++++++++++++++++---------- > 1 file changed, 17 insertions(+), 10 deletions(-) > > diff --git a/drivers/phy/rockchip/phy-rockchip-emmc.c b/drivers/phy/rockchip/phy-rockchip-emmc.c > index f1b24f1..547b746 100644 > --- a/drivers/phy/rockchip/phy-rockchip-emmc.c > +++ b/drivers/phy/rockchip/phy-rockchip-emmc.c > @@ -76,6 +76,10 @@ > #define PHYCTRL_OTAPDLYSEL_MASK 0xf > #define PHYCTRL_OTAPDLYSEL_SHIFT 0x7 > > +#define PHYCTRL_IS_CALDONE(x) \ > + ((((x) >> PHYCTRL_CALDONE_SHIFT) & \ > + PHYCTRL_CALDONE_MASK) == PHYCTRL_CALDONE_DONE) > + > struct rockchip_emmc_phy { > unsigned int reg_offset; > struct regmap *reg_base; > @@ -90,6 +94,7 @@ static int rockchip_emmc_phy_power(struct phy *phy, bool on_off) > unsigned int freqsel = PHYCTRL_FREQSEL_200M; > unsigned long rate; > unsigned long timeout; > + int ret; > > /* > * Keep phyctrl_pdb and phyctrl_endll low to allow > @@ -160,17 +165,19 @@ static int rockchip_emmc_phy_power(struct phy *phy, bool on_off) > PHYCTRL_PDB_SHIFT)); > > /* > - * According to the user manual, it asks driver to > - * wait 5us for calpad busy trimming > + * According to the user manual, it asks driver to wait 5us for > + * calpad busy trimming. However it is documented that this value is > + * PVT(A.K.A process,voltage and temperature) relevant, so some > + * failure cases are found which indicates we should be more tolerant > + * to calpad busy trimming. > */ > - udelay(5); > - regmap_read(rk_phy->reg_base, > - rk_phy->reg_offset + GRF_EMMCPHY_STATUS, > - &caldone); > - caldone = (caldone >> PHYCTRL_CALDONE_SHIFT) & PHYCTRL_CALDONE_MASK; > - if (caldone != PHYCTRL_CALDONE_DONE) { > - pr_err("rockchip_emmc_phy_power: caldone timeout.\n"); > - return -ETIMEDOUT; > + ret = regmap_read_poll_timeout(rk_phy->reg_base, > + rk_phy->reg_offset + GRF_EMMCPHY_STATUS, > + caldone, PHYCTRL_IS_CALDONE(caldone), > + 5, 50); See comments in part 2 of this series, AKA <https://patchwork.kernel.org/patch/10154797/>. I think this should be "0, 50", not "5, 50". ...or, if you insist, "10, 50" > + if (ret) { > + pr_err("%s: caldone failed %d.\n", __func__, ret); I like this v2 slightly better than Caesar's v2 because you changed the word "timeout" to "failed".
diff --git a/drivers/phy/rockchip/phy-rockchip-emmc.c b/drivers/phy/rockchip/phy-rockchip-emmc.c index f1b24f1..547b746 100644 --- a/drivers/phy/rockchip/phy-rockchip-emmc.c +++ b/drivers/phy/rockchip/phy-rockchip-emmc.c @@ -76,6 +76,10 @@ #define PHYCTRL_OTAPDLYSEL_MASK 0xf #define PHYCTRL_OTAPDLYSEL_SHIFT 0x7 +#define PHYCTRL_IS_CALDONE(x) \ + ((((x) >> PHYCTRL_CALDONE_SHIFT) & \ + PHYCTRL_CALDONE_MASK) == PHYCTRL_CALDONE_DONE) + struct rockchip_emmc_phy { unsigned int reg_offset; struct regmap *reg_base; @@ -90,6 +94,7 @@ static int rockchip_emmc_phy_power(struct phy *phy, bool on_off) unsigned int freqsel = PHYCTRL_FREQSEL_200M; unsigned long rate; unsigned long timeout; + int ret; /* * Keep phyctrl_pdb and phyctrl_endll low to allow @@ -160,17 +165,19 @@ static int rockchip_emmc_phy_power(struct phy *phy, bool on_off) PHYCTRL_PDB_SHIFT)); /* - * According to the user manual, it asks driver to - * wait 5us for calpad busy trimming + * According to the user manual, it asks driver to wait 5us for + * calpad busy trimming. However it is documented that this value is + * PVT(A.K.A process,voltage and temperature) relevant, so some + * failure cases are found which indicates we should be more tolerant + * to calpad busy trimming. */ - udelay(5); - regmap_read(rk_phy->reg_base, - rk_phy->reg_offset + GRF_EMMCPHY_STATUS, - &caldone); - caldone = (caldone >> PHYCTRL_CALDONE_SHIFT) & PHYCTRL_CALDONE_MASK; - if (caldone != PHYCTRL_CALDONE_DONE) { - pr_err("rockchip_emmc_phy_power: caldone timeout.\n"); - return -ETIMEDOUT; + ret = regmap_read_poll_timeout(rk_phy->reg_base, + rk_phy->reg_offset + GRF_EMMCPHY_STATUS, + caldone, PHYCTRL_IS_CALDONE(caldone), + 5, 50); + if (ret) { + pr_err("%s: caldone failed %d.\n", __func__, ret); + return ret; } /* Set the frequency of the DLL operation */