Hi Wolfram,
On Tue, Nov 19, 2019 at 9:47 PM Wolfram Sang <wsa@xxxxxxxxxxxxx> wrote:
> > Interestingly, this patch has been part of renesas-drivers since the
> > 2019-10-15-v5.4-rc3 release, without real issues.
>
> Huh, interesting. With which branch does this appear then? linux-next?
renesas-drivers
I can reproduce this with renesas-drivers-2019-11-19-v5.4-rc8 and
renesas_defconfig by booting a ramdisk, and reading from /dev/mmcblk1.
Also with renesas-drivers-2019-10-15-v5.4-rc3.
So some of my local code on top must have impacted the behavior.
> > Today, it was fairly reproducible, so I managed to bisect it to commit
> > 7a7dab237027939c ("mmc: tmio: remove workaround for NON_REMOVABLE") in
> > mmc/next. Reverting this commit fixes the issue.
>
> Hmm, probably we should do the revert despite our discusstion here. And
> then resend the original patch after we figured the cause of this hang.
>
> > The issue can also be fixed by:
> > 1. enabling the hs400_4taps and/or hs400_disabled quirks in
> > sdhi_quirks_match[], OR
> > 2. forcing use_4tap = true in renesas_sdhi_check_scc_error().
> >
> > Salvator-X(S) with R-Car H3 ES1.0 & ES2.0, or M3-W ES1.0, the issue
> > does not show up (probably because of sdhi_quirks_match[]).
> >
> > Do you have a clue?
>
> Not very clear. M3-N is not a 4tap-device, so this can't be a fix.
> However, both disabling HS400 as well as using 4tap will prevent the SCC
> error checking in renesas_sdhi_check_scc_error(). I'd assume the SCC
> hangs.
Makes sense.
> I am working on an issue where the SCC hangs, but this has to do with
> always providing the SCC clock (SDnH). I don't really see the connection
> of that to RuntimePM yet, though :/
Makes sense: this is consistent with the behavior when accessing
registers without enabling the corresponding module clock: it hangs.
So this can happen with other clocks, too.
One more reason not to delegate clock handling to a guest, as doing it
wrong can take down the host, too...
> Can you test this simple workaround patch instead of the revert just so
> we get an idea if these issues are related?
Thanks, applying your workaround on top of
renesas-drivers-2019-11-19-v5.4-rc8 fixes the issue.
This fix is part of renesas/topic/sdhi-manual-calib, right?
And thus has been present in some renesas-drivers release, but was
dropped _before_ the 2019-10-15-v5.4-rc3 release.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
