:: Re: [maemo-leste] [PATCH] Revert "m…
Forside
Slet denne besked
Besvar denne besked
Skribent: Pali Rohár
Dato:  
Til: Faiz Abbas
CC: Peng Fan, u-boot@lists.denx.de, maemo-leste@lists.dyne.org
Nye-emner: Re: [maemo-leste] [PATCH] Revert "mmc: disable UHS modes if Vcc cannot be switched on and off"
Emne: Re: [maemo-leste] [PATCH] Revert "mmc: disable UHS modes if Vcc cannot be switched on and off"
On Tuesday 11 August 2020 18:27:23 Pali Rohár wrote:
> On Tuesday 11 August 2020 21:33:19 Faiz Abbas wrote:
> > Pali,
> >
> > On 11/08/20 1:19 pm, Pali Rohár wrote:
> > > On Tuesday 11 August 2020 08:39:24 Faiz Abbas wrote:
> > >> Pali, Peng,
> > >>
> > >> On 10/08/20 6:25 am, Peng Fan wrote:
> > >>> Faiz, Jean
> > >>>
> > >>>> Subject: [PATCH] Revert "mmc: disable UHS modes if Vcc cannot be switched
> > >>>> on and off"
> > >>>
> > >>> Got time to take a look?
> > >>>
> > >>
> > >> When this issue was reported in the last thread, Pali said that he was unable to get
> > >> prints because of the constant reboot loops? Shouldn't it be easy to put while(1)
> > >> at different points in this execution and see which step causes the reboot loop?
> > >
> > > In May in that last thread I wrote details which I was able to gather:
> > >
> > >     Month ago I was able to detect that problem is somewhere in function
> > >     mmc_set_ios() from mmc.c file. I put long debug line prior and also
> > >     after mmc_set_ios() call. And it was printed only prior. Not after.
> > >     So I think NULL pointer dereference is in mmc_set_ios() function.

> > >
> > > I could try with that while(1) loop to print detailed log message and
> > > read/capture it. But what information for debugging you need? Or what do
> > > you want to print which could help you?
> > >
> >
> > You can continue to bisect into omap_hsmmc_set_ios() to see at what point the NULL pointer
> > gets triggered and what the NULL pointer is.
>
> I could try it, but I do not think I would be able to gather more data.
> I will try to find free time for this debugging on device either at the
> end of this week or end of next week.


Here are more details. Crash is in omap_hsmmc_stop_clock() function when
trying to dereference mmc_base->sysctl.

Call trace is:

mmc_set_ios --> mmc->cfg->ops->set_ios --> omap_hsmmc_set_ios --> omap_hsmmc_stop_clock

I applied following diff

diff --git a/drivers/mmc/omap_hsmmc.c b/drivers/mmc/omap_hsmmc.c
index 715eee0e3e..d4bbfd7b97 100644
--- a/drivers/mmc/omap_hsmmc.c
+++ b/drivers/mmc/omap_hsmmc.c
@@ -22,6 +22,8 @@
* MA 02111-1307 USA
*/

+#define DEBUG
+
 #include <config.h>
 #include <common.h>
 #include <cpu_func.h>
@@ -1335,7 +1337,16 @@ static int mmc_write_data(struct hsmmc *mmc_base, const char *buf,
 #endif
 static void omap_hsmmc_stop_clock(struct hsmmc *mmc_base)
 {
+    debug("omap_hsmmc_stop_clock: before writel\n");
+    debug("mmc_base=%p\n", mmc_base);
+    debug("barrier\n");
+    asm volatile ("");
+    debug("mmc_base->sysctl=%x\n", mmc_base->sysctl);
+    debug("barrier\n");
+    asm volatile ("");
+    debug("readl(&mmc_base->sysctl)=%x\n", readl(&mmc_base->sysctl));
     writel(readl(&mmc_base->sysctl) & ~CEN_ENABLE, &mmc_base->sysctl);
+    debug("omap_hsmmc_stop_clock: after writel\n");
 }


static void omap_hsmmc_start_clock(struct hsmmc *mmc_base)

and I see that first word "barrier" is written. There is no
"mmc_base->sysctl=" string on output display.

If you are interested, mmc_base has value 480b4000.

That is probably all what I can do.

> As I wrote in previous thread, main issue which makes it hard to debug
> is that this error is not triggered in emulator.
>
> > Also can you point to your config file?
>
> I'm using standard config file without any modifications. It is:
> configs/nokia_rx51_defconfig
>
> In case you are interested, I'm compiling u-boot by commands:
>
> export ARCH=arm
> export CROSS_COMPILE=arm-linux-gnueabi-
> make nokia_rx51_config
> make -j12 u-boot.bin