| Message ID | 1421171497-9440-1-git-send-email-dianders@chromium.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi, On Tue, Jan 13, 2015 at 9:51 AM, Doug Anderson <dianders@chromium.org> wrote: > We've seen at least one card that can get confused during all the > errors generated during tuning on rk3288. These errors seem to go > away with DW_MCI_QUIRK_RETRY_DELAY. That quirk is documented to be > for 2.11a and on rk3288 I believe we have 2.70a, but it seems to be > the same type of problem. > > Specifically it looks like in certain error conditions (I saw this > with Response CRC errors) that data keeps showing up in the FIFO even > after the error is reported and the CD (command done) bit is set. If > we don't wait for this data to finish transferring then it confuses > the next transaction. In the specific failure case I ran into I found > that I could monitor the data_state_mc_busy bit and wait for it to > clear, but in other failure cases this bit was stuck at busy when we > saw an error. Hence a generic big delay seems like the only option. > > I have seen this on rk3288, but adding it as a quirk for all Rockchip > IP for now since most previous quirks have shown up on older IP as > well. > > Signed-off-by: Doug Anderson <dianders@chromium.org> > --- > NOTES: > * This is based atop Addy's patch (mmc: dw_mmc: add quirk for broken > data transfer over scheme) at > <https://patchwork.kernel.org/patch/5565431/>, but it would be trivial > to apply this patch without that one if that were desirable. > * If someone has a better solution to this problem, I'm all ears. > > drivers/mmc/host/dw_mmc-rockchip.c | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/drivers/mmc/host/dw_mmc-rockchip.c b/drivers/mmc/host/dw_mmc-rockchip.c > index ba92ebd..b4caf1a 100644 > --- a/drivers/mmc/host/dw_mmc-rockchip.c > +++ b/drivers/mmc/host/dw_mmc-rockchip.c > @@ -74,7 +74,8 @@ static int dw_mci_rockchip_init(struct dw_mci *host) > host->sdio_id0 = 8; > > /* It needs this quirk on all Rockchip SoCs */ > - host->pdata->quirks |= DW_MCI_QUIRK_BROKEN_DTO; > + host->pdata->quirks |= DW_MCI_QUIRK_BROKEN_DTO | > + DW_MCI_QUIRK_RETRY_DELAY; > > return 0; > } All of my recent testing shows that with all of the other fixes Addy and I have posted recently that this one is no longer needed. I think it can be considered Abandoned. -Doug
diff --git a/drivers/mmc/host/dw_mmc-rockchip.c b/drivers/mmc/host/dw_mmc-rockchip.c index ba92ebd..b4caf1a 100644 --- a/drivers/mmc/host/dw_mmc-rockchip.c +++ b/drivers/mmc/host/dw_mmc-rockchip.c @@ -74,7 +74,8 @@ static int dw_mci_rockchip_init(struct dw_mci *host) host->sdio_id0 = 8; /* It needs this quirk on all Rockchip SoCs */ - host->pdata->quirks |= DW_MCI_QUIRK_BROKEN_DTO; + host->pdata->quirks |= DW_MCI_QUIRK_BROKEN_DTO | + DW_MCI_QUIRK_RETRY_DELAY; return 0; }
We've seen at least one card that can get confused during all the errors generated during tuning on rk3288. These errors seem to go away with DW_MCI_QUIRK_RETRY_DELAY. That quirk is documented to be for 2.11a and on rk3288 I believe we have 2.70a, but it seems to be the same type of problem. Specifically it looks like in certain error conditions (I saw this with Response CRC errors) that data keeps showing up in the FIFO even after the error is reported and the CD (command done) bit is set. If we don't wait for this data to finish transferring then it confuses the next transaction. In the specific failure case I ran into I found that I could monitor the data_state_mc_busy bit and wait for it to clear, but in other failure cases this bit was stuck at busy when we saw an error. Hence a generic big delay seems like the only option. I have seen this on rk3288, but adding it as a quirk for all Rockchip IP for now since most previous quirks have shown up on older IP as well. Signed-off-by: Doug Anderson <dianders@chromium.org> --- NOTES: * This is based atop Addy's patch (mmc: dw_mmc: add quirk for broken data transfer over scheme) at <https://patchwork.kernel.org/patch/5565431/>, but it would be trivial to apply this patch without that one if that were desirable. * If someone has a better solution to this problem, I'm all ears. drivers/mmc/host/dw_mmc-rockchip.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)